Config schema reference

A curated guide to the fields operators change most often, followed by the complete list of keys, generated from the Pydantic TRWConfig model. Run trw-mcp config-reference for the same fields with their environment variables.

File location

The config file lives at .trw/config.yaml at the repo root. It is created by trw-mcp init-project. Initialization writes a small baseline plus selected targets and overrides; unspecified fields inherit the remaining defaults.

Precedence (highest wins):

Environment variables prefixed TRW_.
.trw/config.yaml overrides.
Machine defaults in ~/.trw/config.yaml.
Field defaults declared in models/config/_fields_*.py.

Unknown keys (in both env and config.yaml) are ignored - model_config.extra = "ignore" on the BaseSettings class. A retired key is ignored too, with a warning that names it.

Top-level keys

The flat keys you see in config.yaml are the same as the field names on TRWConfig. A curated list of the keys users touch most often:

Key: framework_version
Type: str
Default: "v27.6_TRW"
Description: Pinned framework version for this repo.

Key: target_platforms
Type: list[str]
Default: ["claude-code"]
Description: Selected client targets. Dedicated integrations sync native surfaces; the first entry resolves the primary profile.

Key: ceremony_mode
Type: "full" | "light"
Default: "full"
Description: Profile-derived when unchanged. light suppresses chatty nudges.

Key: response_format
Type: "yaml" | "json"
Default: "yaml"
Description: MCP tool response format.

Key: installation_id
Type: str
Default: ""
Description: Installation ID used for telemetry correlation. Hashed (non-reversible double SHA-256) before transmission — the raw value never leaves the machine.

Key: parallelism_max
Type: int
Default: 10
Description: Maximum concurrent shards or waves during orchestration.

Key: timebox_hours
Type: int
Default: 8
Description: Total session timebox in hours.

Key: task_root
Type: str
Default: "docs"
Description: Root directory for PRDs and task artifacts.

Key: runs_root
Type: str
Default: ".trw/runs"
Description: Directory for run artifacts.

Key: build_check_enabled
Type: bool
Default: true
Description: Whether recorded build-check evidence participates in delivery gating.

Key: embeddings_enabled
Type: bool
Default: true
Description: Dense vector embeddings on by default (needs the trw-memory[embeddings] embedder, which the installer adds unless --no-embeddings).

Key: platform_url
Type: str
Default: ""
Description: Primary platform URL for sync and telemetry.

Key: platform_api_key
Type: SecretStr
Default: ""
Description: Deprecated config-file credential. Use TRW_PLATFORM_API_KEY or .trw/credentials.yaml.

Key	Type	Default	Description
framework_version	str	"v27.6_TRW"	Pinned framework version for this repo.
target_platforms	list[str]	["claude-code"]	Selected client targets. Dedicated integrations sync native surfaces; the first entry resolves the primary profile.
ceremony_mode	"full" \| "light"	"full"	Profile-derived when unchanged. light suppresses chatty nudges.
response_format	"yaml" \| "json"	"yaml"	MCP tool response format.
installation_id	str	""	Installation ID used for telemetry correlation. Hashed (non-reversible double SHA-256) before transmission — the raw value never leaves the machine.
parallelism_max	int	10	Maximum concurrent shards or waves during orchestration.
timebox_hours	int	8	Total session timebox in hours.
task_root	str	"docs"	Root directory for PRDs and task artifacts.
runs_root	str	".trw/runs"	Directory for run artifacts.
build_check_enabled	bool	true	Whether recorded build-check evidence participates in delivery gating.
embeddings_enabled	bool	true	Dense vector embeddings on by default (needs the trw-memory[embeddings] embedder, which the installer adds unless --no-embeddings).
platform_url	str	""	Primary platform URL for sync and telemetry.
platform_api_key	SecretStr	""	Deprecated config-file credential. Use TRW_PLATFORM_API_KEY or .trw/credentials.yaml.

Check source for the hundreds of fields in the complete model: src/trw_mcp/models/config/. Run trw-mcp config-reference for a generated list keyed by env var name.

Curated field groups

These tables group related flat fields for reading; their headings are not all runtime property names. Actual type-narrowed projections include config.build, config.memory, config.orchestration, and config.tools. Ceremony controls remain flat or surface through config.ceremony_feedback and config.surfaces. Projection models are defined in _sub_models.py. At the YAML layer, the listed settings remain flat keys.

build

Key: build_check_enabled
Default: true
Description: Require recorded build-check evidence at the delivery gate; does not run tests.

Key: build_check_coverage_min
Default: 85.0
Description: Minimum coverage percentage to pass the gate.

Key: build_gate_enforcement
Default: "lenient"
Description: One of strict, lenient, off.

Key	Default	Description
build_check_enabled	true	Require recorded build-check evidence at the delivery gate; does not run tests.
build_check_coverage_min	85.0	Minimum coverage percentage to pass the gate.
build_gate_enforcement	"lenient"	One of strict, lenient, off.

memory

Key: learning_max_entries
Default: 500
Description: Cap on total learnings before pruning runs.

Key: recall_max_results
Default: 25
Description: Default result cap for trw_recall.

Key: dedup_skip_threshold
Default: 0.95
Description: Cosine threshold above which dedupe treats entries as duplicates.

Key: dedup_merge_threshold
Default: 0.85
Description: Cosine threshold above which near-duplicates merge.

Key: memory_consolidation_enabled
Default: true
Description: Periodic cluster consolidation of related entries.

Key: recall_user_tier_cap
Default: 5
Description: Max user:local hits allowed into one recall result, so machine-wide entries cannot bury project precision.

Key	Default	Description
learning_max_entries	500	Cap on total learnings before pruning runs.
recall_max_results	25	Default result cap for trw_recall.
dedup_skip_threshold	0.95	Cosine threshold above which dedupe treats entries as duplicates.
dedup_merge_threshold	0.85	Cosine threshold above which near-duplicates merge.
memory_consolidation_enabled	true	Periodic cluster consolidation of related entries.
recall_user_tier_cap	5	Max user:local hits allowed into one recall result, so machine-wide entries cannot bury project precision.

orchestration

Key: parallelism_max
Default: 10
Description: Maximum parallel shards.

Key: timebox_hours
Default: 8
Description: Session timebox.

Key: auto_recall_enabled
Default: true
Description: Allow contextual recall in the trw_session_start pipeline; not per-tool injection.

Key	Default	Description
parallelism_max	10	Maximum parallel shards.
timebox_hours	8	Session timebox.
auto_recall_enabled	true	Allow contextual recall in the trw_session_start pipeline; not per-tool injection.

ceremony fields

Key: nudge_enabled
Default: null
Description: Null defers to profile. Set explicit true or false to force.

Key: phase_gate_enforcement
Default: "lenient"
Description: strict, lenient, or off.

Key: cleanup_on_boot
Default: true
Description: Run the stale-pin and stale-run sweep on server startup.

Key: run_staleness_hours
Default: 48
Description: Hours with no activity before a run is sweep-eligible.

Key	Default	Description
nudge_enabled	null	Null defers to profile. Set explicit true or false to force.
phase_gate_enforcement	"lenient"	strict, lenient, or off.
cleanup_on_boot	true	Run the stale-pin and stale-run sweep on server startup.
run_staleness_hours	48	Hours with no activity before a run is sweep-eligible.

tools projection

Key: tool_resolution_mode
Default: "standard"
Description: standard (default) or all. The surface is flat either way: every session sees the always-on tools plus each config-gated tool whose flag (comms_enabled, dispatch_tools_exposed, assess_enabled) is on. all additionally turns on the comms tools and trw_assess, never trw_dispatch.

Key	Default	Description
tool_resolution_mode	"standard"	standard (default) or all. The surface is flat either way: every session sees the always-on tools plus each config-gated tool whose flag (comms_enabled, dispatch_tools_exposed, assess_enabled) is on. all additionally turns on the comms tools and trw_assess, never trw_dispatch.

Example config

Trimmed and commented example of .trw/config.yaml. Every field not listed falls back through the machine layer and then its built-in default.

# .trw/config.yaml
framework_version: v27.6_TRW

# Orchestration
parallelism_max: 10
timebox_hours: 8

# Platform sync (optional - local runs work without these)
platform_urls:
  - https://api.trwframework.com
  - http://localhost:8000
platform_telemetry_enabled: true
installation_id: trw-framework-dev

# Client profiles to keep in sync.
target_platforms:
  - claude-code
  - cursor-ide
  - cursor-cli
  - opencode
  - codex
  - copilot
  - antigravity-cli

# Memory
embeddings_enabled: true

# Secrets belong in TRW_PLATFORM_API_KEY or .trw/credentials.yaml.

Generate your own with trw-mcp init-project; it writes the baseline keys plus your selected targets and overrides.

Env var overrides

Flat fields on TRWConfig accept environment-variable overrides with the TRW_ prefix. The lookup is case-insensitive; unknown vars are ignored (the same policy used for unknown YAML keys).

Pattern: Scalar
Example variable: TRW_TASK_ROOT=docs
Maps to config key: task_root

Pattern: Number
Example variable: TRW_BUILD_CHECK_COVERAGE_MIN=90
Maps to config key: build_check_coverage_min

Pattern: Bool
Example variable: TRW_EMBEDDINGS_ENABLED=true
Maps to config key: embeddings_enabled

Pattern: List (JSON)
Example variable: TRW_TARGET_PLATFORMS=["cursor-ide","codex"]
Maps to config key: target_platforms

Pattern: Secret
Example variable: TRW_PLATFORM_API_KEY=trw_dk_…
Maps to config key: platform_api_key

Pattern	Example variable	Maps to config key
Scalar	TRW_TASK_ROOT=docs	task_root
Number	TRW_BUILD_CHECK_COVERAGE_MIN=90	build_check_coverage_min
Bool	TRW_EMBEDDINGS_ENABLED=true	embeddings_enabled
List (JSON)	TRW_TARGET_PLATFORMS=["cursor-ide","codex"]	target_platforms
Secret	TRW_PLATFORM_API_KEY=trw_dk_…	platform_api_key

Generator for the full list: trw-mcp config-reference. Prefix configured in _main_fields.py via SettingsConfigDict(env_prefix="TRW_").

All keys

Every one of the 310 keys TRWConfig reads, generated from the model on each release, so this list cannot fall behind the code.

Show all 310 keys

Key: aaref_version
Type: str
Default: "v3.3.0"
Description: AARE-F requirements-framework version stamp deployed with the project.

Key: access_count_utility_boost_cap
Type: float
Default: 0.15
Description: Largest utility bonus a learning can earn from being recalled often.

Key: additional_repo_roots
Type: list[str]
Default: computed
Description: Sibling repository roots for multi-repo workspaces, so PRD key-file references into sibling repos resolve instead of failing as missing paths.

Key: agent_report_max_chars
Type: int
Default: 800
Description: Character limit each bundled agent is asked to keep its final report within (guidance to the model); 0 renders no limit.

Key: agents_enabled
Type: bool | None
Default: null
Description: Install bundled sub-agent definitions; unset follows the client profile.

Key: agents_md_enabled
Type: bool
Default: true
Description: Write TRW's block into `AGENTS.md`.

Key: ambiguity_rate_max
Type: float
Default: 0.05
Description: Highest allowed share of ambiguous terms in a PRD (0-1).

Key: anchor_validity_verified_floor
Type: float
Default: 1.0
Description: Minimum recomputed anchor_validity for a 'verified' verification verdict.

Key: assertion_failure_penalty
Type: float
Default: 0.15
Description: Recall-rank penalty (0-1) for a learning whose stored assertions fail verification, scaled by the failing fraction.

Key: assertion_stale_threshold_days
Type: int
Default: 30
Description: Days after which a learning's assertion verification is stale and due for re-verification.

Key: assess_enabled
Type: bool
Default: false
Description: Expose the assess_support capability pack (trw_assess) on the resolved tool surface.

Key: auto_checkpoint_pre_compact
Type: bool
Default: true
Description: Write a checkpoint automatically before the client compacts context.

Key: auto_recall_enabled
Type: bool
Default: true
Description: Let the prompt hook inject relevant learnings before each prompt (the master `learning_recall_enabled` switch outranks it).

Key: auto_recall_max_results
Type: int
Default: 3
Description: Most learnings the prompt hook injects per prompt.

Key: auto_recall_max_tokens
Type: int
Default: 100
Description: Token budget for the learnings the prompt hook injects per prompt.

Key: auto_recall_min_score
Type: float
Default: 0.35
Description: Minimum relevance (0-1, IDF-weighted share of prompt keywords a learning covers) a learning needs to be injected by the prompt hook.

Key: auto_recall_scan_cap
Type: int
Default: 10000
Description: Most stored learnings the prompt hook's auto-recall scorer scans per prompt (0 or a small value scores only a recent window).

Key: backend_api_key
Type: str
Default: ""
Description: API key for the sync backend; empty means none configured.

Key: backend_url
Type: str
Default: ""
Description: Base URL of the sync backend; empty means no backend configured.

Key: backup_remote_enabled
Type: bool
Default: false
Description: Allow uploading a backup archive of the local store to the platform; off by default because the archive is the raw local store.

Key: boot_deferred_work_budget_ms
Type: int
Default: 5000
Description: Budget (ms) for the post-'initialize' backend-sync resolution scheduled by PRD-CORE-248 FR01.

Key: boot_gc_deferred
Type: bool
Default: true
Description: When True, the boot-time stale-run + stale-pin sweep runs in a daemon background thread ('trw-boot-gc') instead of synchronously before the MCP initialize handshake, so large repos do not stall client connect (production feedback sub_psVs_nUWnLJGvOs3).

Key: build_check_coverage_min
Type: float
Default: 85.0
Description: Minimum test coverage percentage a build check must report to pass the build gate.

Key: build_check_enabled
Type: bool
Default: true
Description: Run the build-check gate; false turns the build-check requirement off for validation and deliver.

Key: build_gate_enforcement
Type: Literal['strict', 'lenient', 'off']
Default: "lenient"
Description: How the build gate reacts to a failed check: `strict` blocks, `lenient` warns, `off` ignores it.

Key: ceremony_feedback_escalation_threshold
Type: float
Default: 60.0
Description: Ceremony score below which recent sessions escalate the ceremony level.

Key: ceremony_feedback_escalation_window
Type: int
Default: 5
Description: Number of recent sessions considered when deciding to escalate ceremony.

Key: ceremony_feedback_min_samples
Type: int
Default: 10
Description: Sessions needed before ceremony feedback adjusts the ceremony level.

Key: ceremony_feedback_quality_threshold
Type: float
Default: 0.9
Description: Outcome quality (0-1) at or above which a session counts as good for feedback.

Key: ceremony_feedback_score_threshold
Type: float
Default: 80.0
Description: Ceremony score at or above which a session counts as well-run for feedback.

Key: ceremony_mode
Type: Literal['full', 'light']
Default: "full"
Description: Ceremony weight: `full` or `light` (shorter guidance and fewer nudges).

Key: changelog_advisory_enabled
Type: bool
Default: false
Description: Warn at deliver when source changed but the package changelog did not; off by default.

Key: checkpoint_suggest_hours
Type: int
Default: 4
Description: Heartbeat-age threshold (hours) at which trw_checkpoint(heartbeat=True) reports should_checkpoint=True to the caller.

Key: cleanup_on_boot
Type: bool
Default: true
Description: When True, the MCP server runs the stale-pin + stale-run sweep on startup.

Key: code_index_bounds
Type: CodeIndexBounds
Default: computed
Description: PRD-CORE-300-FR15 build and query budgets for the local code index: directory entries, files, source bytes and wall-clock per build; query characters and terms, candidate rows, response bytes and wall-clock per query.

Key: code_index_exclude_dirs
Type: list[str]
Default: computed
Description: Directory names the code index skips when scanning the project.

Key: code_index_include_extensions
Type: list[str]
Default: computed
Description: File extensions the code index includes when scanning the project.

Key: code_index_max_file_bytes
Type: int
Default: 1000000
Description: Largest file the code index reads.

Key: comms_body_max_bytes
Type: int
Default: 8192
Description: Maximum comms message body size, in UTF-8 bytes.

Key: comms_candidate_ttl_seconds
Type: int
Default: 3600
Description: Seconds an announced comms candidate stays admissible.

Key: comms_enabled
Type: bool
Default: true
Description: Enable the swarm comms tools; they do nothing until a formation is joined.

Key: comms_fetch_max_items
Type: int
Default: 16
Description: Maximum messages returned by a single comms fetch page.

Key: comms_group_body_budget_bytes
Type: int
Default: 16777216
Description: Live message-body bytes one comms group may hold.

Key: comms_group_row_limit
Type: int
Default: 4096
Description: Lifetime rows (messages incl.

Key: comms_lease_ttl_seconds
Type: int
Default: 120
Description: Comms endpoint lease lifetime, in seconds.

Key: comms_message_ttl_seconds
Type: int
Default: 86400
Description: Seconds an unacknowledged comms message stays deliverable.

Key: comms_poll_interval_seconds
Type: int
Default: 15
Description: Advertised comms poll interval, in seconds.

Key: comms_recipient_outstanding_limit
Type: int
Default: 64
Description: Maximum unacknowledged messages outstanding for one recipient.

Key: comms_response_max_bytes
Type: int
Default: 65536
Description: Maximum canonical JSON comms tool response, in UTF-8 bytes, excluding MCP framing.

Key: comms_retry_grace_seconds
Type: int
Default: 3600
Description: Seconds a terminal message body is kept past its deadline.

Key: comms_scope_max_bytes
Type: int
Default: 1024
Description: Maximum comms scope path length, in UTF-8 bytes.

Key: comms_scope_max_recipients
Type: int
Default: 4
Description: Maximum peers one scoped comms notify may reach.

Key: comms_sender_admissions_per_minute
Type: int
Default: 32
Description: Committed admissions allowed per sender per 60 seconds.

Key: comms_sqlite_busy_timeout_ms
Type: int
Default: 5000
Description: Bounded SQLite busy timeout for comms transactions, in ms.

Key: comms_wait_interval_ms
Type: int
Default: 1000
Description: Sleep between bounded inbox wait attempts, in milliseconds.

Key: comms_wait_max_seconds
Type: int
Default: 30
Description: Maximum bounded inbox wait a caller may request, in seconds; 0 disables.

Key: compact_instructions_template
Type: str
Default: ""
Description: Custom text for the instructions preserved across context compaction; empty uses the built-in template.

Key: completeness_min
Type: float
Default: 0.85
Description: Lowest allowed PRD completeness score (0-1).

Key: complexity_hard_override_threshold
Type: int
Default: 2
Description: Number of active risk signals at which the run is forced to the highest complexity tier regardless of score.

Key: complexity_tier_comprehensive
Type: int
Default: 6
Description: Complexity score at or above (this value plus one) that classifies a run as COMPREHENSIVE.

Key: complexity_tier_minimal
Type: int
Default: 1
Description: Complexity score at or below which a run is classified MINIMAL.

Key: complexity_weight_architecture_change
Type: int
Default: 3
Description: Complexity points added when the task changes architecture.

Key: complexity_weight_cross_cutting
Type: int
Default: 2
Description: Complexity points added when the task is cross-cutting.

Key: complexity_weight_external_integration
Type: int
Default: 2
Description: Complexity points added when the task integrates an external system.

Key: complexity_weight_files_affected_max
Type: int
Default: 5
Description: Cap on the complexity points contributed by the number of files a task affects.

Key: complexity_weight_large_refactoring
Type: int
Default: 1
Description: Complexity points added when the task is a large refactoring.

Key: complexity_weight_novel_patterns
Type: int
Default: 3
Description: Complexity points added when the task introduces novel patterns.

Key: compliance_changelog_filename
Type: str
Default: "CHANGELOG.md"
Description: File name of the project changelog the deliver changelog advisory looks at (default `CHANGELOG.md`).

Key: compliance_dir
Type: str
Default: "compliance"
Description: Directory under the TRW state dir that holds compliance records (default `compliance`).

Key: compliance_review_retention_days
Type: int
Default: 365
Description: Days automated review records are kept before they may be pruned.

Key: context_dir
Type: str
Default: "context"
Description: Directory under the TRW state dir that holds live session context files (default `context`).

Key: cross_model_provider
Type: str
Default: ""
Description: Dispatch client id (for example `codex`) used for cross-model review; empty means no reviewer.

Key: cross_model_review_enabled
Type: bool
Default: false
Description: Ask a second model family for a review of the diff in addition to the local review; off by default.

Key: cross_model_review_timeout_secs
Type: int
Default: 30
Description: Seconds to wait for the cross-model reviewer before giving up.

Key: ctx_isolation_enabled
Type: bool
Default: true
Description: Master kill-switch for per-connection pin isolation.

Key: debug
Type: bool
Default: false
Description: Enable DEBUG-level logging and the .trw/logs/ file sink for every MCP client

Key: dedup_enabled
Type: bool
Default: true
Description: Check a new learning against existing ones and skip or merge near-duplicates.

Key: dedup_merge_threshold
Type: float
Default: 0.85
Description: Similarity (0-1) at or above which a new learning is merged into an existing one.

Key: dedup_skip_threshold
Type: float
Default: 0.95
Description: Similarity (0-1) at or above which a new learning is skipped as a duplicate.

Key: deferred_batch_max_seconds
Type: int
Default: 300
Description: Seconds the whole deferred post-deliver batch may run.

Key: deferred_lock_stale_seconds
Type: int
Default: 600
Description: Seconds after which a deferred-delivery lock is treated as abandoned and taken over.

Key: deferred_step_max_seconds
Type: int
Default: 60
Description: Seconds one deferred post-deliver maintenance step may run before it is abandoned.

Key: degenerate_result_cooldown_calls
Type: int
Default: 20
Description: Matching tool results to skip after an advisory fires, per session (NFR06 noise budget).

Key: degenerate_result_deadline_ms
Type: int
Default: computed
Description: Self-imposed wall-clock deadline; past it the adapter exits 0 without emitting (NFR01).

Key: degenerate_result_freshness_commands
Type: list[str]
Default: computed
Description: Command prefixes whose undated output cannot distinguish 'nothing there' from 'could not look'.

Key: degenerate_result_max_read_bytes
Type: int
Default: 65536
Description: Cap on bytes read from a PostToolUse tool_response before shape classification (NFR03).

Key: degenerate_result_truncation_markers
Type: list[str]
Default: computed
Description: Literal markers whose presence means the result was capped rather than complete.

Key: degraded_detect_grace_seconds
Type: int
Default: 180
Description: Seconds after the SessionStart epoch marker before the absent-MCP-surface detector may conclude the surface is absent.

Key: degraded_detect_min_prompts
Type: int
Default: 2
Description: Minimum prompt index in this session before the absent-MCP-surface detector may fire.

Key: degraded_event_tail_lines
Type: int
Default: 500
Description: PRD-FIX-128 external audit row 7.

Key: degraded_marker_retention_hours
Type: int
Default: 24
Description: How long an epoch marker or emission latch belonging to an identity with NO live pin is kept before the SessionStart sweep reclaims it.

Key: deliver_gate_mode
Type: Literal['advisory', 'block_coding', 'block_all']
Default: "block_coding"
Description: Deliver gate posture: `advisory` warns only, `block_coding` blocks a missing build check for coding-type tasks, `block_all` blocks for every task type.

Key: deliver_gate_task_type_overrides
Type: dict[str, str]
Default: computed
Description: Per-task-type override of `deliver_gate_mode`, e.g.

Key: deliver_gate_unclassified_change_threshold
Type: int
Default: 1
Description: Number of distinct files a session must modify before the missing-build deliver gate arms for a task type that does not inherently expect a build.

Key: deliver_graph_backfill_deadline_seconds
Type: float
Default: 2.0
Description: Seconds the deliver-time knowledge-graph backfill may run before it stops.

Key: deliver_graph_backfill_enabled
Type: bool
Default: true
Description: On deliver, build knowledge-graph edges for not-yet-graphed learnings within a small time budget.

Key: delivery_busy_timeout_ms
Type: int
Default: 5000
Description: SQLite busy timeout for the delivery operation store.

Key: delivery_operations_mode
Type: Literal['off', 'observe', 'enforce']
Default: "enforce"
Description: Delivery journal mode: `off` disables it, `observe` records without acting, `enforce` acts on it.

Key: delivery_queue_depth_max
Type: int
Default: 128
Description: Bounded deferred FIFO queue depth (FR06).

Key: delivery_stale_lease_minutes
Type: int
Default: 15
Description: Minutes a pending lease must be stale before FR04 takeover.

Key: dispatch_child_trw_access
Type: bool
Default: false
Description: Default for dispatch --with-trw: inject ONLY TRW's own stdio trw-mcp server into the child's argv so a dispatched peer can use TRW session/memory tools.

Key: dispatch_default_client
Type: str | None
Default: "codex"
Description: Client used when --client is omitted; None requires an explicit --client.

Key: dispatch_default_effort
Type: Literal['low', 'medium', 'high', 'xhigh', 'max'] | None
Default: null
Description: Effort for dispatched children when the request names none; overrides the role's table row.

Key: dispatch_default_efforts
Type: dict[Literal['claude', 'codex', 'agy', 'opencode', 'cursor-cli', 'copilot', 'grok'], Literal['low', 'medium', 'high', 'xhigh', 'max']]
Default: computed
Description: Per-client effort for dispatched children (e.g.

Key: dispatch_default_max_turns
Type: int
Default: 30
Description: Turn cap for dispatched children (clients with a verified flag only); 0 disables it.

Key: dispatch_default_models
Type: dict[str, str]
Default: computed
Description: Per-client model override applied when --model is omitted (e.g.

Key: dispatch_default_read_only
Type: bool
Default: true
Description: Default read-only posture for dispatched children; --allow-writes overrides to False.

Key: dispatch_default_timeout_s
Type: int
Default: 600
Description: Wall-clock timeout (seconds) applied when --timeout is omitted.

Key: dispatch_enabled_clients
Type: list[Literal['claude', 'codex', 'agy', 'opencode', 'cursor-cli', 'copilot', 'grok']]
Default: computed
Description: Dispatch targets the CLI may launch; a client outside this list is rejected.

Key: dispatch_fallback_clients
Type: list[str]
Default: computed
Description: Clients tried in order after quota_exhausted or a launch failure; --fallback-clients overrides.

Key: dispatch_max_concurrent_children
Type: int
Default: 0
Description: Per-user cap on concurrently running dispatch children across processes; 0 disables it.

Key: dispatch_require_effort
Type: bool
Default: false
Description: Refuse (exit 2) a dispatch with no effort from --effort, dispatch_default_effort or the role table when its client and model accept an effort; clients without a carrier pass.

Key: dispatch_slot_wait_s
Type: float
Default: 600.0
Description: Seconds a top-level dispatch waits for a free child slot before refusing with silence_reason concurrency_cap; a nested dispatch never waits.

Key: dispatch_tools_exposed
Type: bool
Default: false
Description: Expose the dispatch capability pack (trw_dispatch: launch, status and the AgentWorkEvidence modes) on the resolved tool surface.

Key: dispatch_version_probe_timeout_s
Type: int
Default: 5
Description: Per-client wall-clock bound (seconds) on the doctor formation-readiness version probe; N enabled clients are bounded by N times this value.

Key: display_timezone
Type: str
Default: "UTC"
Description: IANA timezone used to render local times in `trw_status`; stored and compared times stay UTC.

Key: dry_check_enabled
Type: bool
Default: true
Description: Run the duplicate-code (DRY) check in the build gate.

Key: dry_check_min_block_size
Type: int
Default: 5
Description: Fewest identical lines that count as a duplicated block in the DRY check.

Key: embeddings_coverage_warn_threshold
Type: float
Default: 0.1
Description: Share (0-1) of learnings without embeddings above which the pipeline-health check warns.

Key: embeddings_enabled
Type: bool
Default: true
Description: Governs this trw-mcp process only (dedup, the installer/swap extras, retrieval probes); it does not switch off the shared per-user memory daemon, whose embeddings are set by MEMORY_EMBEDDINGS_ENABLED=false in the daemon's environment (then restart it).

Key: entries_dir
Type: str
Default: "entries"
Description: Directory under the learnings dir that holds individual learning entries (default `entries`).

Key: events_file
Type: str
Default: "events.jsonl"
Description: File name of the per-run event log (default `events.jsonl`).

Key: evidence_receipt_mode
Type: Literal['observe', 'enforce']
Default: "enforce"
Description: Evidence receipt compatibility: `observe` writes typed receipts alongside the legacy files, `enforce` requires the typed receipts.

Key: extra_prd_categories
Type: list[str]
Default: computed
Description: Project-specific PRD category names accepted in addition to the built-in generic set (CORE, QUAL, INFRA, LOCAL, EXPLR, RESEARCH, FIX).

Key: factory_enabled
Type: bool
Default: false
Description: Enable the experimental software-factory commands (status, receipts, factory checkpoints); off unless you opt in.

Key: feedback
Type: FeedbackFields
Default: computed
Description: Settings for the feedback channel behind `trw_status(feedback=...)`.

Key: formation_activity_stall_seconds
Type: int
Default: 1800
Description: Seconds without activity after which a formation member is reported as stalled.

Key: formation_deliver_gate
Type: Literal['block', 'advisory']
Default: "block"
Description: Whether delivering while a joined formation member is mid-work blocks (`block`) or only warns (`advisory`).

Key: formation_hook_ownership_mode
Type: Literal['warn', 'off']
Default: "warn"
Description: PreToolUse ownership check for formation members: `warn` reports an unowned edit, `off` disables it (never blocks).

Key: formation_manifest_lock_timeout_seconds
Type: float
Default: 10.0
Description: Bounded wait for the formation manifest lock, in seconds.

Key: formation_ownership_enforcement
Type: Literal['refuse', 'warn']
Default: "refuse"
Description: Commit-boundary check that a member only commits paths it owns: `refuse` blocks the commit, `warn` only reports it.

Key: formation_status_member_limit
Type: int
Default: 16
Description: Maximum member rows assembled by the formation status roll-up.

Key: framework_md_enabled
Type: bool | None
Default: null
Description: Deploy the framework document into the project; unset follows the client profile.

Key: framework_read_scope
Type: Literal['phase', 'full']
Default: "phase"
Description: Whether the SessionStart framework read directive names only the sections for the current phase ('phase', the default, at most 6349 characters) or the whole 35073-character document ('full').

Key: framework_version
Type: str
Default: "v27.6_TRW"
Description: Framework version stamp deployed with the project (managed by `make framework-bump`, not by hand).

Key: frameworks_dir
Type: str
Default: "frameworks"
Description: Directory under the TRW state dir where the deployed framework documents live (default `frameworks`).

Key: hint_delivery_canary_window
Type: int
Default: 200
Description: Newest N pre-edit hint records the hint_delivery doctor row inspects (20-5000).

Key: hint_delivery_fallback_warn_share
Type: float
Default: 0.5
Description: Fallback-status share of the hint_delivery window at/above which the row WARNs (0-1).

Key: hint_delivery_receipt_stale_days
Type: int
Default: 7
Description: Post-commit sidecar receipt age (days) after which hint_delivery WARNs when trw-distill is installed.

Key: hint_hub_downrank
Type: bool
Default: true
Description: On a hub file (more anchored lessons than `hint_hub_threshold`), list anchor-only lessons after the text-matched ones; false restores anchored-first ordering.

Key: hint_hub_threshold
Type: int
Default: 67
Description: Anchored lessons above which a file is a hub whose anchor-only lessons the pre-edit hint shows after text-matched ones (1-9999; lower flags more files, each edit fetches this + 1 rows).

Key: hint_recall_deadline_ms
Type: int
Default: 600
Description: Wall-clock budget (ms) for the pre-edit hint's T1 learnings recall (50-10000); past it the hint returns no learnings as recall_timeout rather than block the T2 sidecar hint.

Key: hint_sidecar_ancestor_enabled
Type: bool
Default: true
Description: Let the pre-edit hint use a proven-ancestor commit's sidecar when the exact-HEAD one is missing, instead of no sidecar half at all.

Key: hint_sidecar_max_commits_behind
Type: int
Default: 500
Description: Commits an ancestor sidecar may trail HEAD by and still serve the pre-edit hint (1-10000; lower refuses older sidecars as sidecar_too_far_behind).

Key: hint_sidecar_rebuild_after_commits
Type: int
Default: 150
Description: Commits behind HEAD at which a served ancestor sidecar triggers a rebuild request (1-10000, capped at hint_sidecar_max_commits_behind; lower rebuilds more often).

Key: hint_sidecar_rebuild_min_interval_minutes
Type: int
Default: 15
Description: Minimum minutes between two detached sidecar rebuild requests (1-1440; lower starts builds more often, higher leaves the sidecar older).

Key: hint_sidecar_refresh_enabled
Type: bool
Default: true
Description: Post-commit hook and session start request a detached rebuild of the trw-distill pre-edit-hint sidecar (post-commit also refreshes the risk report); false stops both.

Key: hooks_enabled
Type: bool
Default: true
Description: Master switch for every bundled hook; false disables them all.

Key: hybrid_rrf_k
Type: int
Default: 60
Description: Reciprocal-rank-fusion constant used to merge keyword and vector recall rankings.

Key: index_auto_sync_on_status_change
Type: bool
Default: true
Description: Regenerate the PRD index automatically when a PRD's status changes.

Key: installation_id
Type: str
Default: ""
Description: Explicit installation identifier; empty derives a stable `inst-<hash>` from the project root path.

Key: installer_answers
Type: dict[str, bool]
Default: computed
Description: Answers the installer remembers so it never asks twice: key -> true/false (for example ai_extras, proprietary_upgrade, migrate_learnings, platform_connect, system_python_<id>).

Key: instruction_backup_dir
Type: str
Default: ".trw/backups/instructions"
Description: Project-relative directory where instruction-file backups are written before a guarded rewrite.

Key: instruction_backup_retention
Type: int
Default: 5
Description: Pre-write backup copies retained per instruction filename (PRD-FIX-123-FR04).

Key: instruction_catalogue_mode
Type: Literal['pointer', 'verbatim']
Default: "pointer"
Description: Whether a full-ceremony client's generated protocol block carries a pointer to trw_status(detail="surface") ('pointer', the default) or the verbatim tool table ('verbatim').

Key: instruction_dry_run_diff_max_lines
Type: int
Default: 400
Description: Maximum unified-diff lines returned per target by a dry-run instruction sync (PRD-FIX-123-FR03).

Key: instruction_size_gate_mode
Type: Literal['warn', 'block'] | None
Default: null
Description: Whether an over-size TRW instruction block only warns (`warn`) or is refused (`block`); unset lets the client profile decide.

Key: instruction_write_block_delta_tolerance_bytes
Type: int
Default: 64
Description: Bytes of slack when testing whether a total-size drop is explained by the TRW block's own size change (PRD-FIX-123-FR02).

Key: instruction_write_max_total_shrink_fraction
Type: float
Default: 0.25
Description: Largest fraction of an instruction file's total bytes a guarded write may remove before it is refused without force (PRD-FIX-123-FR02).

Key: intel_cache_enabled
Type: bool
Default: true
Description: Cache fetched intelligence payloads locally for `intel_cache_ttl_seconds`.

Key: intel_cache_ttl_seconds
Type: int
Default: 3600
Description: Seconds a fetched intelligence payload stays cached before it is refetched.

Key: knowledge_jaccard_threshold
Type: float
Default: 0.3
Description: Tag-overlap (Jaccard) similarity at which two learnings are linked in the knowledge topology.

Key: knowledge_min_cluster_size
Type: int
Default: 3
Description: Fewest learnings a topic cluster needs to be written out.

Key: knowledge_output_dir
Type: str
Default: "knowledge"
Description: Directory under the TRW state dir where knowledge-topology pages are written (default `knowledge`).

Key: knowledge_sync_threshold
Type: int
Default: 50
Description: Number of new learnings that triggers a knowledge-topology rebuild.

Key: learn_journal_drain_budget_ms
Type: int
Default: 3000
Description: Milliseconds a session-start learn-journal sweep may spend replaying before it stops and leaves the rest.

Key: learn_journal_drain_limit
Type: int
Default: 50
Description: Most pending learn-journal records replayed per session-start sweep; the rest drain on later sweeps.

Key: learn_journal_enabled
Type: bool
Default: true
Description: Durably journal an accepted `trw_learn` before the slow store pipeline so a crash mid-write loses nothing; replayed at the next session start.

Key: learn_journal_max_replay_attempts
Type: int
Default: 5
Description: Replay attempts before a record that keeps failing transiently is dead-lettered instead of retried forever.

Key: learning_auto_prune_cap
Type: int
Default: 150
Description: Active learnings kept by the auto-prune pass; lowest-value ones beyond it are pruned.

Key: learning_auto_prune_max_seconds
Type: int
Default: 30
Description: Wall-clock budget in seconds for one auto-prune pass; a pass that hits it returns the partial result.

Key: learning_auto_prune_min_interval_hours
Type: int
Default: 24
Description: Least hours between two auto-prune passes (a pass is expensive on large stores).

Key: learning_auto_prune_on_deliver
Type: bool
Default: true
Description: Prune low-value learnings automatically after deliver.

Key: learning_decay_half_life_days
Type: float
Default: 14.0
Description: Days for a learning's utility to halve when unused.

Key: learning_decay_use_exponent
Type: float
Default: 0.6
Description: Exponent applied to use count when slowing a learning's decay.

Key: learning_max_entries
Type: int
Default: 500
Description: Most learnings kept in the store; the lowest-value ones beyond it are dropped by maintenance.

Key: learning_outcome_correlation_scope
Type: str
Default: "session"
Description: Which recalls an outcome is attributed to: `session` or a wider scope.

Key: learning_outcome_correlation_window_minutes
Type: int
Default: 7
Description: Minutes after a recall within which a build outcome is attributed to the learnings shown.

Key: learning_recall_enabled
Type: bool
Default: true
Description: Master switch for learnings reaching the agent through recall, hints and hooks; false keeps them all off.

Key: learning_repeated_op_threshold
Type: int
Default: 3
Description: Times the same operation must recur before it is suggested as a learning.

Key: learning_sharing_enabled
Type: bool
Default: false
Description: Separate consent to publish learning content (summary and detail) to the platform; off by default and distinct from usage telemetry.

Key: learning_utility_delete_threshold
Type: float
Default: 0.05
Description: Utility below which a learning is deleted outright.

Key: learning_utility_prune_threshold
Type: float
Default: 0.1
Description: Utility below which a learning becomes a prune candidate.

Key: learnings_dir
Type: str
Default: "learnings"
Description: Directory under the TRW state dir that holds learnings (default `learnings`).

Key: llm_default_model
Type: str
Default: "haiku"
Description: Model alias used for TRW's own small LLM calls (default `haiku`).

Key: llm_usage_log_enabled
Type: bool
Default: true
Description: Log every TRW LLM call's usage to `llm_usage_log_file`.

Key: llm_usage_log_file
Type: str
Default: "llm_usage.jsonl"
Description: File name of the LLM usage log (default `llm_usage.jsonl`).

Key: logs_dir
Type: str
Default: "logs"
Description: Directory under the TRW state dir that holds log and usage files (default `logs`).

Key: maintain_verify_batch_limit
Type: int
Default: 1000
Description: Max entries per maintain-verify page; the explicit sweep traverses all eligible pages.

Key: max_auto_lines
Type: int
Default: 300
Description: Line limit for the whole instruction file after TRW's block is merged in; a write that would exceed it is refused, never truncated.

Key: max_consolidated_tags
Type: int
Default: 20
Description: Most tags kept on a consolidated (merged) learning.

Key: memory_consolidation_enabled
Type: bool
Default: true
Description: Run periodic consolidation that merges clusters of similar learnings.

Key: memory_consolidation_max_per_cycle
Type: int
Default: 50
Description: Most learnings consolidated in one cycle.

Key: memory_consolidation_min_cluster
Type: int
Default: 3
Description: Fewest similar learnings that form a cluster worth consolidating.

Key: memory_consolidation_similarity_threshold
Type: float
Default: 0.75
Description: Similarity (0-1) two learnings need to be clustered for consolidation.

Key: meta_tune
Type: MetaTuneConfig
Default: computed
Description: Nested meta-tuning settings (`meta_tune.enabled` mirrors `meta_tune_enabled`).

Key: meta_tune_enabled
Type: bool
Default: false
Description: Enable the meta-tuning loop (self-modification sandbox); off by default and only supported on Linux.

Key: migration_gate_enabled
Type: bool
Default: true
Description: Run the database-migration check in the build gate.

Key: model_family
Type: str
Default: ""
Description: Model family tag for surface telemetry and nudge logs (C-5).

Key: nudge_density
Type: Literal['low', 'medium', 'high'] | None
Default: null
Description: Nudge injection density: low=less frequent, high=more frequent; None defers to profile default.

Key: nudge_enabled
Type: bool | None
Default: null
Description: Show ceremony nudges to the agent; unset follows the client profile.

Key: nudge_messenger
Type: Literal['standard', 'minimal', 'contextual', 'contextual_action', 'contextual_distress', 'silent_flow', 'cod', 'stepback', 'anchor', 'negative', 'governance'] | None
Default: null
Description: Nudge wording variant (`standard`, `minimal`, `contextual`, ...); unset follows the client profile.

Key: nudge_pool_cooldown_after
Type: int
Default: 3
Description: Times a pooled nudge may show before it goes on cooldown.

Key: nudge_pool_cooldown_calls
Type: int
Default: 10
Description: Tool calls a pooled nudge stays silent once it is on cooldown.

Key: nudge_pool_cooldown_wall_clock_max_hours
Type: int
Default: 24
Description: Max wall-clock hours a nudge pool stays cooled down before re-engaging.

Key: nudge_pool_weights
Type: models._client_profile.NudgePoolWeights | None
Default: null
Description: Project override for nudge pool routing weights (must sum to 100); None defers to profile.

Key: nudge_variant
Type: str | None
Default: null
Description: Label stamped on nudge events to compare arms in live traffic; routing still follows `nudge_messenger`.

Key: otel_enabled
Type: bool
Default: false
Description: Set up OpenTelemetry tracing at startup; off by default.

Key: parallelism_max
Type: int
Default: 10
Description: Parallelism ceiling written into a new project's `config.yaml` at init (default 10).

Key: path_index_exclude_dirs
Type: list[str]
Default: computed
Description: Extra directory names pruned from the PRD path-index walk, unioned with the built-in PATH_INDEX_EXCLUDE_DIRS.

Key: path_index_max_files
Type: int
Default: 1000000
Description: RUNAWAY GUARD (not the latency control) on files scanned while building the bare-filename basename index.

Key: path_index_max_seconds
Type: float
Default: 5.0
Description: RUNAWAY GUARD (wall-clock seconds) on the bare-filename basename index walk.

Key: patterns_dir
Type: str
Default: "patterns"
Description: Directory under the TRW state dir that holds recorded patterns (default `patterns`).

Key: phase_gate_enforcement
Type: Literal['strict', 'lenient', 'off']
Default: "lenient"
Description: How phase-exit gates react to unmet criteria: `strict` blocks, `lenient` warns, `off` ignores.

Key: pin_ttl_hours
Type: int
Default: 24
Description: Time-to-live (hours) for entries in the persistent pin store before GC evicts them.

Key: pipeline_health_gate_enabled
Type: bool
Default: true
Description: Master kill switch for the FR06 fail-closed pipeline-health gate.

Key: pipeline_health_gate_failure_threshold
Type: int
Default: 10
Description: Consecutive push failures before the FR06 gate fails closed.

Key: pipeline_health_gate_graph_min_corpus
Type: int
Default: 50
Description: Young-store threshold: an empty knowledge graph fails the FR06 gate, and low embedding coverage degrades the pipeline, only for a store holding more memories than this.

Key: pipeline_health_gate_stale_hours
Type: float
Default: 6.0
Description: Hours since last successful push before the FR06 gate fails closed.

Key: platform_api_key
Type: SecretStr
Default: ""
Description: API key for the TRW platform (kept secret in output); empty means none configured.

Key: platform_contact_enabled
Type: bool
Default: true
Description: Master veto on every network call to the TRW platform (sync, backup, telemetry uploads); false keeps the install fully offline.

Key: platform_telemetry_enabled
Type: bool
Default: false
Description: Send anonymous usage telemetry to the TRW platform; off by default and separate from learning sharing.

Key: platform_url
Type: str
Default: ""
Description: Base URL of the TRW platform this install talks to; empty means none configured.

Key: platform_urls
Type: list[str]
Default: computed
Description: Ordered list of platform URLs; the first is used and the rest are fallbacks.

Key: post_commit_distill_incremental
Type: bool
Default: false
Description: Let the post-commit hook run `trw-distill run --incremental --live-ingest`, which may call an LLM and writes lessons into the store; off unless you choose it.

Key: prd_min_content_density
Type: float
Default: 0.3
Description: Least content density a PRD needs to move from DRAFT to REVIEW.

Key: prd_required_status_for_implement
Type: str
Default: "approved"
Description: PRD status required before the implement phase may start (default `approved`).

Key: prd_transition_gate
Type: Literal['warn', 'block']
Default: "block"
Description: Whether moving a PRD to implemented without accepted evidence only warns (`warn`) or is blocked (`block`).

Key: prd_validate_budget_seconds
Type: float
Default: 60.0
Description: Wall-clock budget (seconds) for trw_prd_validate's dynamic checks; on breach the remaining check groups are skipped and the result is flagged validation_partial (PRD-FIX-112).

Key: prd_validation_cache_maintenance_interval
Type: int
Default: 32
Description: Number of successful shard writes between bounded maintenance sweeps.

Key: prd_validation_cache_max_entries
Type: int
Default: 512
Description: Maximum number of cached pure-validation shards.

Key: prd_validation_cache_max_entry_bytes
Type: int
Default: 4194304
Description: Maximum size of a single cached shard.

Key: prd_validation_cache_max_total_bytes
Type: int
Default: 67108864
Description: Maximum aggregate on-disk bytes across all cached pure-validation shards.

Key: prds_relative_path
Type: str
Default: "docs/requirements-aare-f/prds"
Description: Project-relative directory that holds PRD documents (default `docs/requirements-aare-f/prds`).

Key: pricing_table_path
Type: str
Default: ""
Description: Path to a custom model pricing table for cost estimates; empty uses the bundled one.

Key: profile_domain_path_map
Type: dict[str, str]
Default: computed
Description: Source-path prefix -> profile 'domain' layer name, used by infer_domain when no explicit domain is supplied.

Key: profile_system_enabled
Type: bool
Default: true
Description: Resolve and apply the layered client/domain profile at session start; false skips profile resolution.

Key: project_handoff_path
Type: str
Default: ".trw/HANDOFF.md"
Description: Project-relative path of the deliver-time handoff file (PRD-CORE-249-FR01).

Key: project_namespace
Type: str
Default: ""
Description: Stable memory-store namespace for this project once migrated to the machine-local daemon store; empty means the checkout keeps its own `.trw/memory` store.

Key: protection_tier_prune_discount
Type: dict[str, float]
Default: computed
Description: Multiplier on the utility threshold an entry must fall below before an automatic prune, tier demotion or purge may nominate it (PRD-CORE-244 FR10).

Key: recall_internal_fields
Type: frozenset[str]
Default: ["access_count", "anchor_validity", "combined_score", "last_accessed_at", "outcome_history", "recall_count", "recurrence", "session_count"]
Description: Learning fields removed from recall results because they are internal.

Key: recall_max_results
Type: int
Default: 25
Description: Most learnings `trw_recall` returns for one query.

Key: recall_provenance_inline
Type: bool
Default: true
Description: Show source and scope inline in each recalled learning stub so provenance is checked before a claim is reused.

Key: recall_user_tier_cap
Type: int
Default: 5
Description: Maximum number of user-tier hits that may enter a single federated recall result, so a flood of low-value user hits cannot bury a precise project hit (PRD-CORE-185 FR06 / D3 / R4).

Key: recall_utility_lambda
Type: float
Default: 0.3
Description: Weight of utility in targeted recall ranking; 0 disables utility preferences (query relevance always wins).

Key: receipts_dir
Type: str
Default: "receipts"
Description: Directory under the TRW state dir that holds build/review receipts (default `receipts`).

Key: reflect_max_success_patterns
Type: int
Default: 5
Description: Most success patterns kept per reflection.

Key: reflections_dir
Type: str
Default: "reflections"
Description: Directory under the TRW state dir that holds reflection outputs (default `reflections`).

Key: requirement_drift_gate
Type: Literal['warn', 'block'] | None
Default: null
Description: Deliver-time requirement-drift gate: `warn` or `block`; unset keeps the per-PRD default (block for safety-critical PRDs, warn otherwise).

Key: response_format
Type: Literal['yaml', 'json']
Default: "yaml"
Description: Format of tool responses: `yaml` or `json`.

Key: review_confidence_threshold
Type: int
Default: 80
Description: Minimum finding confidence (0-100) for an automated review finding to be reported.

Key: review_gate_mode
Type: Literal['warn', 'block']
Default: "warn"
Description: Whether a STANDARD or COMPREHENSIVE run that reaches deliver with no recorded review only warns (`warn`) or is blocked (`block`).

Key: review_mandate_advisory_enabled
Type: bool
Default: true
Description: Tell the agent up front that STANDARD and COMPREHENSIVE runs need a review before deliver (advisory only).

Key: review_signoff_ttl_hours
Type: int
Default: 24
Description: Max validity window of an operator review sign-off, in hours.

Key: review_verdict_ttl_hours
Type: int
Default: 24
Description: Hours a recorded review verdict stays valid at deliver time; older ones are treated as expired even if the reviewed bytes did not move.

Key: risk_scaling_enabled
Type: bool
Default: true
Description: Scale PRD validation thresholds by the PRD's risk level.

Key: run_auto_close_enabled
Type: bool
Default: true
Description: Close stale runs automatically when run status is assembled.

Key: run_stale_ttl_hours
Type: int
Default: 48
Description: Hours without activity after which a run counts as stale and may be auto-closed.

Key: run_staleness_grace_hours
Type: int
Default: 12
Description: Grace window (hours) appended to run_staleness_hours; runs inside this window emit run_near_stale_warning but are not abandoned.

Key: run_staleness_hours
Type: int
Default: 48
Description: Age (hours) beyond which an active run with no activity is eligible for the stale-run sweep.

Key: runs_root
Type: str
Default: ".trw/runs"
Description: Project-relative directory where run directories are created (default `.trw/runs`).

Key: scoring_default_days_unused
Type: int
Default: 30
Description: Days-unused value assumed for a learning with no recorded last use.

Key: scripts_dir
Type: str
Default: "scripts"
Description: Directory under the TRW state dir that holds generated helper scripts (default `scripts`).

Key: security
Type: SecurityConfig
Default: computed
Description: Nested security settings for the MCP trust boundary.

Key: semantic_checks_enabled
Type: bool
Default: true
Description: Run semantic checks in the build gate in addition to syntax checks.

Key: session_start_recall_enabled
Type: bool | None
Default: null
Description: Recall learnings at session start; unset follows `learning_recall_enabled`.

Key: shared_mcp
Type: SharedMcpConfig
Default: computed
Description: Opt-in shared MCP server settings: `enabled` makes clients launch `trw-mcp-proxy` against one shared server instead of one stdio server each.

Key: skills_enabled
Type: bool | None
Default: null
Description: Install bundled skills; unset follows the client profile.

Key: source_human_utility_boost
Type: float
Default: 0.1
Description: Utility bonus for learnings a human wrote or confirmed.

Key: state_learning_default_ttl_days
Type: dict[str, int]
Default: computed
Description: Default validity window in days, per learning type, proposed for a state-asserting learning (PRD-CORE-244 FR05).

Key: sub_claude_md_max_lines
Type: int
Default: 50
Description: Line limit for a module-level (sub-scope) instruction file.

Key: surface_role
Type: Literal['agent', 'reviewer']
Default: "agent"
Description: MCP surface role: `agent` is the normal full surface, `reviewer` a read-only report surface for dispatched review sessions (normally set by the environment, not here).

Key: sync_health_failure_threshold
Type: int
Default: 10
Description: Consecutive push failures before sync_health marks degraded.

Key: sync_health_stale_hours
Type: float
Default: 6.0
Description: Hours since last successful push before sync_health marks degraded.

Key: sync_interval_seconds
Type: int
Default: 300
Description: Seconds between team-sync push/pull cycles.

Key: sync_pull_timeout_seconds
Type: float
Default: 5.0
Description: Seconds to wait for a pull request before giving up.

Key: sync_push_batch_size
Type: int
Default: 100
Description: Most learnings pushed to the backend in one batch.

Key: sync_push_timeout_seconds
Type: float
Default: 10.0
Description: Seconds to wait for a push request before giving up.

Key: target_platforms
Type: list[str]
Default: computed
Description: Platforms to sync instruction files for.

Key: task_root
Type: str
Default: "docs"
Description: Project-relative directory that holds task directories (default `docs`).

Key: team_sync_all_projects
Type: bool
Default: false
Description: Rank other projects' team learnings like unrecorded ones in recall instead of lower (default: lower).

Key: team_sync_enabled
Type: bool
Default: false
Description: Enable team sync of learnings with the backend; off by default.

Key: team_sync_fresh_after_seconds
Type: int
Default: 300
Description: Pull team learnings before recall when the last pull is older than this (0: never).

Key: telemetry
Type: bool
Default: false
Description: Legacy usage-telemetry switch copied into a new project's `config.yaml` at init; the live switch is `telemetry_enabled`.

Key: telemetry_enabled
Type: bool
Default: true
Description: Record local tool-call telemetry under the logs directory.

Key: telemetry_file
Type: str
Default: "tool-telemetry.jsonl"
Description: File name of the local tool-call telemetry log (default `tool-telemetry.jsonl`).

Key: telemetry_log_max_bytes
Type: int
Default: 10485760
Description: Pipeline telemetry-log rotation threshold in bytes (FR04)

Key: templates_dir
Type: str
Default: "templates"
Description: Directory under the TRW state dir that holds generated templates (default `templates`).

Key: time_tracking_enabled
Type: bool
Default: true
Description: Track elapsed time and ETAs for task-scaled runs and show them in `trw_status`, checkpoints and deliver.

Key: timebox_hours
Type: int
Default: 8
Description: Default time box in hours written into a new project's `config.yaml` at init (default 8).

Key: tool_output_size_warning_bytes
Type: int
Default: 8192
Description: Rendered PostToolUse tool_response byte length above which a size advisory fires (PRD-INFRA-194-FR04).

Key: tool_resolution_mode
Type: Literal['standard', 'all']
Default: "standard"
Description: Which tool packs the MCP server exposes: `standard` is the kernel plus every pack whose flag is on, `all` also enables the comms and assess packs.

Key: traceability_coverage_min
Type: float
Default: 0.9
Description: Lowest allowed share of requirements traced to code or tests (0-1).

Key: transition_nudges_enabled
Type: bool
Default: true
Description: Allow phase-transition nudge lines (learning and suggestion lines) to reach the agent; independent of `nudge_enabled`.

Key: trust_crawl_boundary
Type: int
Default: 50
Description: Session count up to which an agent stays in the `crawl` trust tier; above it the tier is `walk`.

Key: trust_walk_boundary
Type: int
Default: 200
Description: Session count up to which an agent stays in the `walk` trust tier; above it the tier is `run`.

Key: trw_dir
Type: str
Default: ".trw"
Description: Name of the TRW state directory inside the project (default `.trw`).

Key: validation_density_weight
Type: float
Default: 20.0
Description: Points a PRD can earn for content density in the validation score.

Key: validation_draft_threshold
Type: float
Default: 60.0
Description: Validation score at or above which a PRD is at least REVIEW quality.

Key: validation_implementation_readiness_weight
Type: float
Default: 25.0
Description: Points a PRD can earn for implementation readiness in the validation score.

Key: validation_review_threshold
Type: float
Default: 85.0
Description: Validation score at or above which a PRD is APPROVED quality.

Key: validation_skeleton_threshold
Type: float
Default: 30.0
Description: Validation score at or above which a PRD is at least DRAFT quality (below it: skeleton).

Key: validation_structure_weight
Type: float
Default: 20.0
Description: Points a PRD can earn for structure in the validation score.

Key: validation_traceability_weight
Type: float
Default: 35.0
Description: Points a PRD can earn for traceability in the validation score.

Key: verification_cache_ttl_seconds
Type: int
Default: 3600
Description: Freshness window for last-known evidence; never proves current-tree validity or triggers verification.

Key: version_check_interval_seconds
Type: float
Default: 300.0
Description: Interval (seconds) between cached booted-vs-installed trw-mcp version-drift checks on the tool-call middleware path (PRD-CORE-215-FR02).

Key: wal_checkpoint_max_age_seconds
Type: int
Default: 3600
Description: Seconds since the last SUCCESSFUL checkpoint after which one is due regardless of WAL size.

Key: wal_checkpoint_threshold_mb
Type: int
Default: 10
Description: WAL size (MiB) at or above which a checkpoint is due.

Key: wiring_gate_mode
Type: Literal['warn', 'block']
Default: "warn"
Description: Whether a PRD whose requirements are not wired to code only warns (`warn`) or fails validation (`block`).

Key: wiring_gate_mode_overrides
Type: dict[str, Literal['warn', 'block']]
Default: computed
Description: Per-PRD override of `wiring_gate_mode` (PRD id or category to `warn`/`block`); empty means every PRD uses the global mode.

Key	Type	Default	Description
aaref_version	str	"v3.3.0"	AARE-F requirements-framework version stamp deployed with the project.
access_count_utility_boost_cap	float	0.15	Largest utility bonus a learning can earn from being recalled often.
additional_repo_roots	list[str]	computed	Sibling repository roots for multi-repo workspaces, so PRD key-file references into sibling repos resolve instead of failing as missing paths.
agent_report_max_chars	int	800	Character limit each bundled agent is asked to keep its final report within (guidance to the model); 0 renders no limit.
agents_enabled	bool \| None	null	Install bundled sub-agent definitions; unset follows the client profile.
agents_md_enabled	bool	true	Write TRW's block into `AGENTS.md`.
ambiguity_rate_max	float	0.05	Highest allowed share of ambiguous terms in a PRD (0-1).
anchor_validity_verified_floor	float	1.0	Minimum recomputed anchor_validity for a 'verified' verification verdict.
assertion_failure_penalty	float	0.15	Recall-rank penalty (0-1) for a learning whose stored assertions fail verification, scaled by the failing fraction.
assertion_stale_threshold_days	int	30	Days after which a learning's assertion verification is stale and due for re-verification.
assess_enabled	bool	false	Expose the assess_support capability pack (trw_assess) on the resolved tool surface.
auto_checkpoint_pre_compact	bool	true	Write a checkpoint automatically before the client compacts context.
auto_recall_enabled	bool	true	Let the prompt hook inject relevant learnings before each prompt (the master `learning_recall_enabled` switch outranks it).
auto_recall_max_results	int	3	Most learnings the prompt hook injects per prompt.
auto_recall_max_tokens	int	100	Token budget for the learnings the prompt hook injects per prompt.
auto_recall_min_score	float	0.35	Minimum relevance (0-1, IDF-weighted share of prompt keywords a learning covers) a learning needs to be injected by the prompt hook.
auto_recall_scan_cap	int	10000	Most stored learnings the prompt hook's auto-recall scorer scans per prompt (0 or a small value scores only a recent window).
backend_api_key	str	""	API key for the sync backend; empty means none configured.
backend_url	str	""	Base URL of the sync backend; empty means no backend configured.
backup_remote_enabled	bool	false	Allow uploading a backup archive of the local store to the platform; off by default because the archive is the raw local store.
boot_deferred_work_budget_ms	int	5000	Budget (ms) for the post-'initialize' backend-sync resolution scheduled by PRD-CORE-248 FR01.
boot_gc_deferred	bool	true	When True, the boot-time stale-run + stale-pin sweep runs in a daemon background thread ('trw-boot-gc') instead of synchronously before the MCP initialize handshake, so large repos do not stall client connect (production feedback sub_psVs_nUWnLJGvOs3).
build_check_coverage_min	float	85.0	Minimum test coverage percentage a build check must report to pass the build gate.
build_check_enabled	bool	true	Run the build-check gate; false turns the build-check requirement off for validation and deliver.
build_gate_enforcement	Literal['strict', 'lenient', 'off']	"lenient"	How the build gate reacts to a failed check: `strict` blocks, `lenient` warns, `off` ignores it.
ceremony_feedback_escalation_threshold	float	60.0	Ceremony score below which recent sessions escalate the ceremony level.
ceremony_feedback_escalation_window	int	5	Number of recent sessions considered when deciding to escalate ceremony.
ceremony_feedback_min_samples	int	10	Sessions needed before ceremony feedback adjusts the ceremony level.
ceremony_feedback_quality_threshold	float	0.9	Outcome quality (0-1) at or above which a session counts as good for feedback.
ceremony_feedback_score_threshold	float	80.0	Ceremony score at or above which a session counts as well-run for feedback.
ceremony_mode	Literal['full', 'light']	"full"	Ceremony weight: `full` or `light` (shorter guidance and fewer nudges).
changelog_advisory_enabled	bool	false	Warn at deliver when source changed but the package changelog did not; off by default.
checkpoint_suggest_hours	int	4	Heartbeat-age threshold (hours) at which trw_checkpoint(heartbeat=True) reports should_checkpoint=True to the caller.
cleanup_on_boot	bool	true	When True, the MCP server runs the stale-pin + stale-run sweep on startup.
code_index_bounds	CodeIndexBounds	computed	PRD-CORE-300-FR15 build and query budgets for the local code index: directory entries, files, source bytes and wall-clock per build; query characters and terms, candidate rows, response bytes and wall-clock per query.
code_index_exclude_dirs	list[str]	computed	Directory names the code index skips when scanning the project.
code_index_include_extensions	list[str]	computed	File extensions the code index includes when scanning the project.
code_index_max_file_bytes	int	1000000	Largest file the code index reads.
comms_body_max_bytes	int	8192	Maximum comms message body size, in UTF-8 bytes.
comms_candidate_ttl_seconds	int	3600	Seconds an announced comms candidate stays admissible.
comms_enabled	bool	true	Enable the swarm comms tools; they do nothing until a formation is joined.
comms_fetch_max_items	int	16	Maximum messages returned by a single comms fetch page.
comms_group_body_budget_bytes	int	16777216	Live message-body bytes one comms group may hold.
comms_group_row_limit	int	4096	Lifetime rows (messages incl.
comms_lease_ttl_seconds	int	120	Comms endpoint lease lifetime, in seconds.
comms_message_ttl_seconds	int	86400	Seconds an unacknowledged comms message stays deliverable.
comms_poll_interval_seconds	int	15	Advertised comms poll interval, in seconds.
comms_recipient_outstanding_limit	int	64	Maximum unacknowledged messages outstanding for one recipient.
comms_response_max_bytes	int	65536	Maximum canonical JSON comms tool response, in UTF-8 bytes, excluding MCP framing.
comms_retry_grace_seconds	int	3600	Seconds a terminal message body is kept past its deadline.
comms_scope_max_bytes	int	1024	Maximum comms scope path length, in UTF-8 bytes.
comms_scope_max_recipients	int	4	Maximum peers one scoped comms notify may reach.
comms_sender_admissions_per_minute	int	32	Committed admissions allowed per sender per 60 seconds.
comms_sqlite_busy_timeout_ms	int	5000	Bounded SQLite busy timeout for comms transactions, in ms.
comms_wait_interval_ms	int	1000	Sleep between bounded inbox wait attempts, in milliseconds.
comms_wait_max_seconds	int	30	Maximum bounded inbox wait a caller may request, in seconds; 0 disables.
compact_instructions_template	str	""	Custom text for the instructions preserved across context compaction; empty uses the built-in template.
completeness_min	float	0.85	Lowest allowed PRD completeness score (0-1).
complexity_hard_override_threshold	int	2	Number of active risk signals at which the run is forced to the highest complexity tier regardless of score.
complexity_tier_comprehensive	int	6	Complexity score at or above (this value plus one) that classifies a run as COMPREHENSIVE.
complexity_tier_minimal	int	1	Complexity score at or below which a run is classified MINIMAL.
complexity_weight_architecture_change	int	3	Complexity points added when the task changes architecture.
complexity_weight_cross_cutting	int	2	Complexity points added when the task is cross-cutting.
complexity_weight_external_integration	int	2	Complexity points added when the task integrates an external system.
complexity_weight_files_affected_max	int	5	Cap on the complexity points contributed by the number of files a task affects.
complexity_weight_large_refactoring	int	1	Complexity points added when the task is a large refactoring.
complexity_weight_novel_patterns	int	3	Complexity points added when the task introduces novel patterns.
compliance_changelog_filename	str	"CHANGELOG.md"	File name of the project changelog the deliver changelog advisory looks at (default `CHANGELOG.md`).
compliance_dir	str	"compliance"	Directory under the TRW state dir that holds compliance records (default `compliance`).
compliance_review_retention_days	int	365	Days automated review records are kept before they may be pruned.
context_dir	str	"context"	Directory under the TRW state dir that holds live session context files (default `context`).
cross_model_provider	str	""	Dispatch client id (for example `codex`) used for cross-model review; empty means no reviewer.
cross_model_review_enabled	bool	false	Ask a second model family for a review of the diff in addition to the local review; off by default.
cross_model_review_timeout_secs	int	30	Seconds to wait for the cross-model reviewer before giving up.
ctx_isolation_enabled	bool	true	Master kill-switch for per-connection pin isolation.
debug	bool	false	Enable DEBUG-level logging and the .trw/logs/ file sink for every MCP client
dedup_enabled	bool	true	Check a new learning against existing ones and skip or merge near-duplicates.
dedup_merge_threshold	float	0.85	Similarity (0-1) at or above which a new learning is merged into an existing one.
dedup_skip_threshold	float	0.95	Similarity (0-1) at or above which a new learning is skipped as a duplicate.
deferred_batch_max_seconds	int	300	Seconds the whole deferred post-deliver batch may run.
deferred_lock_stale_seconds	int	600	Seconds after which a deferred-delivery lock is treated as abandoned and taken over.
deferred_step_max_seconds	int	60	Seconds one deferred post-deliver maintenance step may run before it is abandoned.
degenerate_result_cooldown_calls	int	20	Matching tool results to skip after an advisory fires, per session (NFR06 noise budget).
degenerate_result_deadline_ms	int	computed	Self-imposed wall-clock deadline; past it the adapter exits 0 without emitting (NFR01).
degenerate_result_freshness_commands	list[str]	computed	Command prefixes whose undated output cannot distinguish 'nothing there' from 'could not look'.
degenerate_result_max_read_bytes	int	65536	Cap on bytes read from a PostToolUse tool_response before shape classification (NFR03).
degenerate_result_truncation_markers	list[str]	computed	Literal markers whose presence means the result was capped rather than complete.
degraded_detect_grace_seconds	int	180	Seconds after the SessionStart epoch marker before the absent-MCP-surface detector may conclude the surface is absent.
degraded_detect_min_prompts	int	2	Minimum prompt index in this session before the absent-MCP-surface detector may fire.
degraded_event_tail_lines	int	500	PRD-FIX-128 external audit row 7.
degraded_marker_retention_hours	int	24	How long an epoch marker or emission latch belonging to an identity with NO live pin is kept before the SessionStart sweep reclaims it.
deliver_gate_mode	Literal['advisory', 'block_coding', 'block_all']	"block_coding"	Deliver gate posture: `advisory` warns only, `block_coding` blocks a missing build check for coding-type tasks, `block_all` blocks for every task type.
deliver_gate_task_type_overrides	dict[str, str]	computed	Per-task-type override of `deliver_gate_mode`, e.g.
deliver_gate_unclassified_change_threshold	int	1	Number of distinct files a session must modify before the missing-build deliver gate arms for a task type that does not inherently expect a build.
deliver_graph_backfill_deadline_seconds	float	2.0	Seconds the deliver-time knowledge-graph backfill may run before it stops.
deliver_graph_backfill_enabled	bool	true	On deliver, build knowledge-graph edges for not-yet-graphed learnings within a small time budget.
delivery_busy_timeout_ms	int	5000	SQLite busy timeout for the delivery operation store.
delivery_operations_mode	Literal['off', 'observe', 'enforce']	"enforce"	Delivery journal mode: `off` disables it, `observe` records without acting, `enforce` acts on it.
delivery_queue_depth_max	int	128	Bounded deferred FIFO queue depth (FR06).
delivery_stale_lease_minutes	int	15	Minutes a pending lease must be stale before FR04 takeover.
dispatch_child_trw_access	bool	false	Default for dispatch --with-trw: inject ONLY TRW's own stdio trw-mcp server into the child's argv so a dispatched peer can use TRW session/memory tools.
dispatch_default_client	str \| None	"codex"	Client used when --client is omitted; None requires an explicit --client.
dispatch_default_effort	Literal['low', 'medium', 'high', 'xhigh', 'max'] \| None	null	Effort for dispatched children when the request names none; overrides the role's table row.
dispatch_default_efforts	dict[Literal['claude', 'codex', 'agy', 'opencode', 'cursor-cli', 'copilot', 'grok'], Literal['low', 'medium', 'high', 'xhigh', 'max']]	computed	Per-client effort for dispatched children (e.g.
dispatch_default_max_turns	int	30	Turn cap for dispatched children (clients with a verified flag only); 0 disables it.
dispatch_default_models	dict[str, str]	computed	Per-client model override applied when --model is omitted (e.g.
dispatch_default_read_only	bool	true	Default read-only posture for dispatched children; --allow-writes overrides to False.
dispatch_default_timeout_s	int	600	Wall-clock timeout (seconds) applied when --timeout is omitted.
dispatch_enabled_clients	list[Literal['claude', 'codex', 'agy', 'opencode', 'cursor-cli', 'copilot', 'grok']]	computed	Dispatch targets the CLI may launch; a client outside this list is rejected.
dispatch_fallback_clients	list[str]	computed	Clients tried in order after quota_exhausted or a launch failure; --fallback-clients overrides.
dispatch_max_concurrent_children	int	0	Per-user cap on concurrently running dispatch children across processes; 0 disables it.
dispatch_require_effort	bool	false	Refuse (exit 2) a dispatch with no effort from --effort, dispatch_default_effort or the role table when its client and model accept an effort; clients without a carrier pass.
dispatch_slot_wait_s	float	600.0	Seconds a top-level dispatch waits for a free child slot before refusing with silence_reason concurrency_cap; a nested dispatch never waits.
dispatch_tools_exposed	bool	false	Expose the dispatch capability pack (trw_dispatch: launch, status and the AgentWorkEvidence modes) on the resolved tool surface.
dispatch_version_probe_timeout_s	int	5	Per-client wall-clock bound (seconds) on the doctor formation-readiness version probe; N enabled clients are bounded by N times this value.
display_timezone	str	"UTC"	IANA timezone used to render local times in `trw_status`; stored and compared times stay UTC.
dry_check_enabled	bool	true	Run the duplicate-code (DRY) check in the build gate.
dry_check_min_block_size	int	5	Fewest identical lines that count as a duplicated block in the DRY check.
embeddings_coverage_warn_threshold	float	0.1	Share (0-1) of learnings without embeddings above which the pipeline-health check warns.
embeddings_enabled	bool	true	Governs this trw-mcp process only (dedup, the installer/swap extras, retrieval probes); it does not switch off the shared per-user memory daemon, whose embeddings are set by MEMORY_EMBEDDINGS_ENABLED=false in the daemon's environment (then restart it).
entries_dir	str	"entries"	Directory under the learnings dir that holds individual learning entries (default `entries`).
events_file	str	"events.jsonl"	File name of the per-run event log (default `events.jsonl`).
evidence_receipt_mode	Literal['observe', 'enforce']	"enforce"	Evidence receipt compatibility: `observe` writes typed receipts alongside the legacy files, `enforce` requires the typed receipts.
extra_prd_categories	list[str]	computed	Project-specific PRD category names accepted in addition to the built-in generic set (CORE, QUAL, INFRA, LOCAL, EXPLR, RESEARCH, FIX).
factory_enabled	bool	false	Enable the experimental software-factory commands (status, receipts, factory checkpoints); off unless you opt in.
feedback	FeedbackFields	computed	Settings for the feedback channel behind `trw_status(feedback=...)`.
formation_activity_stall_seconds	int	1800	Seconds without activity after which a formation member is reported as stalled.
formation_deliver_gate	Literal['block', 'advisory']	"block"	Whether delivering while a joined formation member is mid-work blocks (`block`) or only warns (`advisory`).
formation_hook_ownership_mode	Literal['warn', 'off']	"warn"	PreToolUse ownership check for formation members: `warn` reports an unowned edit, `off` disables it (never blocks).
formation_manifest_lock_timeout_seconds	float	10.0	Bounded wait for the formation manifest lock, in seconds.
formation_ownership_enforcement	Literal['refuse', 'warn']	"refuse"	Commit-boundary check that a member only commits paths it owns: `refuse` blocks the commit, `warn` only reports it.
formation_status_member_limit	int	16	Maximum member rows assembled by the formation status roll-up.
framework_md_enabled	bool \| None	null	Deploy the framework document into the project; unset follows the client profile.
framework_read_scope	Literal['phase', 'full']	"phase"	Whether the SessionStart framework read directive names only the sections for the current phase ('phase', the default, at most 6349 characters) or the whole 35073-character document ('full').
framework_version	str	"v27.6_TRW"	Framework version stamp deployed with the project (managed by `make framework-bump`, not by hand).
frameworks_dir	str	"frameworks"	Directory under the TRW state dir where the deployed framework documents live (default `frameworks`).
hint_delivery_canary_window	int	200	Newest N pre-edit hint records the hint_delivery doctor row inspects (20-5000).
hint_delivery_fallback_warn_share	float	0.5	Fallback-status share of the hint_delivery window at/above which the row WARNs (0-1).
hint_delivery_receipt_stale_days	int	7	Post-commit sidecar receipt age (days) after which hint_delivery WARNs when trw-distill is installed.
hint_hub_downrank	bool	true	On a hub file (more anchored lessons than `hint_hub_threshold`), list anchor-only lessons after the text-matched ones; false restores anchored-first ordering.
hint_hub_threshold	int	67	Anchored lessons above which a file is a hub whose anchor-only lessons the pre-edit hint shows after text-matched ones (1-9999; lower flags more files, each edit fetches this + 1 rows).
hint_recall_deadline_ms	int	600	Wall-clock budget (ms) for the pre-edit hint's T1 learnings recall (50-10000); past it the hint returns no learnings as recall_timeout rather than block the T2 sidecar hint.
hint_sidecar_ancestor_enabled	bool	true	Let the pre-edit hint use a proven-ancestor commit's sidecar when the exact-HEAD one is missing, instead of no sidecar half at all.
hint_sidecar_max_commits_behind	int	500	Commits an ancestor sidecar may trail HEAD by and still serve the pre-edit hint (1-10000; lower refuses older sidecars as sidecar_too_far_behind).
hint_sidecar_rebuild_after_commits	int	150	Commits behind HEAD at which a served ancestor sidecar triggers a rebuild request (1-10000, capped at hint_sidecar_max_commits_behind; lower rebuilds more often).
hint_sidecar_rebuild_min_interval_minutes	int	15	Minimum minutes between two detached sidecar rebuild requests (1-1440; lower starts builds more often, higher leaves the sidecar older).
hint_sidecar_refresh_enabled	bool	true	Post-commit hook and session start request a detached rebuild of the trw-distill pre-edit-hint sidecar (post-commit also refreshes the risk report); false stops both.
hooks_enabled	bool	true	Master switch for every bundled hook; false disables them all.
hybrid_rrf_k	int	60	Reciprocal-rank-fusion constant used to merge keyword and vector recall rankings.
index_auto_sync_on_status_change	bool	true	Regenerate the PRD index automatically when a PRD's status changes.
installation_id	str	""	Explicit installation identifier; empty derives a stable `inst-<hash>` from the project root path.
installer_answers	dict[str, bool]	computed	Answers the installer remembers so it never asks twice: key -> true/false (for example ai_extras, proprietary_upgrade, migrate_learnings, platform_connect, system_python_<id>).
instruction_backup_dir	str	".trw/backups/instructions"	Project-relative directory where instruction-file backups are written before a guarded rewrite.
instruction_backup_retention	int	5	Pre-write backup copies retained per instruction filename (PRD-FIX-123-FR04).
instruction_catalogue_mode	Literal['pointer', 'verbatim']	"pointer"	Whether a full-ceremony client's generated protocol block carries a pointer to trw_status(detail="surface") ('pointer', the default) or the verbatim tool table ('verbatim').
instruction_dry_run_diff_max_lines	int	400	Maximum unified-diff lines returned per target by a dry-run instruction sync (PRD-FIX-123-FR03).
instruction_size_gate_mode	Literal['warn', 'block'] \| None	null	Whether an over-size TRW instruction block only warns (`warn`) or is refused (`block`); unset lets the client profile decide.
instruction_write_block_delta_tolerance_bytes	int	64	Bytes of slack when testing whether a total-size drop is explained by the TRW block's own size change (PRD-FIX-123-FR02).
instruction_write_max_total_shrink_fraction	float	0.25	Largest fraction of an instruction file's total bytes a guarded write may remove before it is refused without force (PRD-FIX-123-FR02).
intel_cache_enabled	bool	true	Cache fetched intelligence payloads locally for `intel_cache_ttl_seconds`.
intel_cache_ttl_seconds	int	3600	Seconds a fetched intelligence payload stays cached before it is refetched.
knowledge_jaccard_threshold	float	0.3	Tag-overlap (Jaccard) similarity at which two learnings are linked in the knowledge topology.
knowledge_min_cluster_size	int	3	Fewest learnings a topic cluster needs to be written out.
knowledge_output_dir	str	"knowledge"	Directory under the TRW state dir where knowledge-topology pages are written (default `knowledge`).
knowledge_sync_threshold	int	50	Number of new learnings that triggers a knowledge-topology rebuild.
learn_journal_drain_budget_ms	int	3000	Milliseconds a session-start learn-journal sweep may spend replaying before it stops and leaves the rest.
learn_journal_drain_limit	int	50	Most pending learn-journal records replayed per session-start sweep; the rest drain on later sweeps.
learn_journal_enabled	bool	true	Durably journal an accepted `trw_learn` before the slow store pipeline so a crash mid-write loses nothing; replayed at the next session start.
learn_journal_max_replay_attempts	int	5	Replay attempts before a record that keeps failing transiently is dead-lettered instead of retried forever.
learning_auto_prune_cap	int	150	Active learnings kept by the auto-prune pass; lowest-value ones beyond it are pruned.
learning_auto_prune_max_seconds	int	30	Wall-clock budget in seconds for one auto-prune pass; a pass that hits it returns the partial result.
learning_auto_prune_min_interval_hours	int	24	Least hours between two auto-prune passes (a pass is expensive on large stores).
learning_auto_prune_on_deliver	bool	true	Prune low-value learnings automatically after deliver.
learning_decay_half_life_days	float	14.0	Days for a learning's utility to halve when unused.
learning_decay_use_exponent	float	0.6	Exponent applied to use count when slowing a learning's decay.
learning_max_entries	int	500	Most learnings kept in the store; the lowest-value ones beyond it are dropped by maintenance.
learning_outcome_correlation_scope	str	"session"	Which recalls an outcome is attributed to: `session` or a wider scope.
learning_outcome_correlation_window_minutes	int	7	Minutes after a recall within which a build outcome is attributed to the learnings shown.
learning_recall_enabled	bool	true	Master switch for learnings reaching the agent through recall, hints and hooks; false keeps them all off.
learning_repeated_op_threshold	int	3	Times the same operation must recur before it is suggested as a learning.
learning_sharing_enabled	bool	false	Separate consent to publish learning content (summary and detail) to the platform; off by default and distinct from usage telemetry.
learning_utility_delete_threshold	float	0.05	Utility below which a learning is deleted outright.
learning_utility_prune_threshold	float	0.1	Utility below which a learning becomes a prune candidate.
learnings_dir	str	"learnings"	Directory under the TRW state dir that holds learnings (default `learnings`).
llm_default_model	str	"haiku"	Model alias used for TRW's own small LLM calls (default `haiku`).
llm_usage_log_enabled	bool	true	Log every TRW LLM call's usage to `llm_usage_log_file`.
llm_usage_log_file	str	"llm_usage.jsonl"	File name of the LLM usage log (default `llm_usage.jsonl`).
logs_dir	str	"logs"	Directory under the TRW state dir that holds log and usage files (default `logs`).
maintain_verify_batch_limit	int	1000	Max entries per maintain-verify page; the explicit sweep traverses all eligible pages.
max_auto_lines	int	300	Line limit for the whole instruction file after TRW's block is merged in; a write that would exceed it is refused, never truncated.
max_consolidated_tags	int	20	Most tags kept on a consolidated (merged) learning.
memory_consolidation_enabled	bool	true	Run periodic consolidation that merges clusters of similar learnings.
memory_consolidation_max_per_cycle	int	50	Most learnings consolidated in one cycle.
memory_consolidation_min_cluster	int	3	Fewest similar learnings that form a cluster worth consolidating.
memory_consolidation_similarity_threshold	float	0.75	Similarity (0-1) two learnings need to be clustered for consolidation.
meta_tune	MetaTuneConfig	computed	Nested meta-tuning settings (`meta_tune.enabled` mirrors `meta_tune_enabled`).
meta_tune_enabled	bool	false	Enable the meta-tuning loop (self-modification sandbox); off by default and only supported on Linux.
migration_gate_enabled	bool	true	Run the database-migration check in the build gate.
model_family	str	""	Model family tag for surface telemetry and nudge logs (C-5).
nudge_density	Literal['low', 'medium', 'high'] \| None	null	Nudge injection density: low=less frequent, high=more frequent; None defers to profile default.
nudge_enabled	bool \| None	null	Show ceremony nudges to the agent; unset follows the client profile.
nudge_messenger	Literal['standard', 'minimal', 'contextual', 'contextual_action', 'contextual_distress', 'silent_flow', 'cod', 'stepback', 'anchor', 'negative', 'governance'] \| None	null	Nudge wording variant (`standard`, `minimal`, `contextual`, ...); unset follows the client profile.
nudge_pool_cooldown_after	int	3	Times a pooled nudge may show before it goes on cooldown.
nudge_pool_cooldown_calls	int	10	Tool calls a pooled nudge stays silent once it is on cooldown.
nudge_pool_cooldown_wall_clock_max_hours	int	24	Max wall-clock hours a nudge pool stays cooled down before re-engaging.
nudge_pool_weights	models._client_profile.NudgePoolWeights \| None	null	Project override for nudge pool routing weights (must sum to 100); None defers to profile.
nudge_variant	str \| None	null	Label stamped on nudge events to compare arms in live traffic; routing still follows `nudge_messenger`.
otel_enabled	bool	false	Set up OpenTelemetry tracing at startup; off by default.
parallelism_max	int	10	Parallelism ceiling written into a new project's `config.yaml` at init (default 10).
path_index_exclude_dirs	list[str]	computed	Extra directory names pruned from the PRD path-index walk, unioned with the built-in PATH_INDEX_EXCLUDE_DIRS.
path_index_max_files	int	1000000	RUNAWAY GUARD (not the latency control) on files scanned while building the bare-filename basename index.
path_index_max_seconds	float	5.0	RUNAWAY GUARD (wall-clock seconds) on the bare-filename basename index walk.
patterns_dir	str	"patterns"	Directory under the TRW state dir that holds recorded patterns (default `patterns`).
phase_gate_enforcement	Literal['strict', 'lenient', 'off']	"lenient"	How phase-exit gates react to unmet criteria: `strict` blocks, `lenient` warns, `off` ignores.
pin_ttl_hours	int	24	Time-to-live (hours) for entries in the persistent pin store before GC evicts them.
pipeline_health_gate_enabled	bool	true	Master kill switch for the FR06 fail-closed pipeline-health gate.
pipeline_health_gate_failure_threshold	int	10	Consecutive push failures before the FR06 gate fails closed.
pipeline_health_gate_graph_min_corpus	int	50	Young-store threshold: an empty knowledge graph fails the FR06 gate, and low embedding coverage degrades the pipeline, only for a store holding more memories than this.
pipeline_health_gate_stale_hours	float	6.0	Hours since last successful push before the FR06 gate fails closed.
platform_api_key	SecretStr	""	API key for the TRW platform (kept secret in output); empty means none configured.
platform_contact_enabled	bool	true	Master veto on every network call to the TRW platform (sync, backup, telemetry uploads); false keeps the install fully offline.
platform_telemetry_enabled	bool	false	Send anonymous usage telemetry to the TRW platform; off by default and separate from learning sharing.
platform_url	str	""	Base URL of the TRW platform this install talks to; empty means none configured.
platform_urls	list[str]	computed	Ordered list of platform URLs; the first is used and the rest are fallbacks.
post_commit_distill_incremental	bool	false	Let the post-commit hook run `trw-distill run --incremental --live-ingest`, which may call an LLM and writes lessons into the store; off unless you choose it.
prd_min_content_density	float	0.3	Least content density a PRD needs to move from DRAFT to REVIEW.
prd_required_status_for_implement	str	"approved"	PRD status required before the implement phase may start (default `approved`).
prd_transition_gate	Literal['warn', 'block']	"block"	Whether moving a PRD to implemented without accepted evidence only warns (`warn`) or is blocked (`block`).
prd_validate_budget_seconds	float	60.0	Wall-clock budget (seconds) for trw_prd_validate's dynamic checks; on breach the remaining check groups are skipped and the result is flagged validation_partial (PRD-FIX-112).
prd_validation_cache_maintenance_interval	int	32	Number of successful shard writes between bounded maintenance sweeps.
prd_validation_cache_max_entries	int	512	Maximum number of cached pure-validation shards.
prd_validation_cache_max_entry_bytes	int	4194304	Maximum size of a single cached shard.
prd_validation_cache_max_total_bytes	int	67108864	Maximum aggregate on-disk bytes across all cached pure-validation shards.
prds_relative_path	str	"docs/requirements-aare-f/prds"	Project-relative directory that holds PRD documents (default `docs/requirements-aare-f/prds`).
pricing_table_path	str	""	Path to a custom model pricing table for cost estimates; empty uses the bundled one.
profile_domain_path_map	dict[str, str]	computed	Source-path prefix -> profile 'domain' layer name, used by infer_domain when no explicit domain is supplied.
profile_system_enabled	bool	true	Resolve and apply the layered client/domain profile at session start; false skips profile resolution.
project_handoff_path	str	".trw/HANDOFF.md"	Project-relative path of the deliver-time handoff file (PRD-CORE-249-FR01).
project_namespace	str	""	Stable memory-store namespace for this project once migrated to the machine-local daemon store; empty means the checkout keeps its own `.trw/memory` store.
protection_tier_prune_discount	dict[str, float]	computed	Multiplier on the utility threshold an entry must fall below before an automatic prune, tier demotion or purge may nominate it (PRD-CORE-244 FR10).
recall_internal_fields	frozenset[str]	["access_count", "anchor_validity", "combined_score", "last_accessed_at", "outcome_history", "recall_count", "recurrence", "session_count"]	Learning fields removed from recall results because they are internal.
recall_max_results	int	25	Most learnings `trw_recall` returns for one query.
recall_provenance_inline	bool	true	Show source and scope inline in each recalled learning stub so provenance is checked before a claim is reused.
recall_user_tier_cap	int	5	Maximum number of user-tier hits that may enter a single federated recall result, so a flood of low-value user hits cannot bury a precise project hit (PRD-CORE-185 FR06 / D3 / R4).
recall_utility_lambda	float	0.3	Weight of utility in targeted recall ranking; 0 disables utility preferences (query relevance always wins).
receipts_dir	str	"receipts"	Directory under the TRW state dir that holds build/review receipts (default `receipts`).
reflect_max_success_patterns	int	5	Most success patterns kept per reflection.
reflections_dir	str	"reflections"	Directory under the TRW state dir that holds reflection outputs (default `reflections`).
requirement_drift_gate	Literal['warn', 'block'] \| None	null	Deliver-time requirement-drift gate: `warn` or `block`; unset keeps the per-PRD default (block for safety-critical PRDs, warn otherwise).
response_format	Literal['yaml', 'json']	"yaml"	Format of tool responses: `yaml` or `json`.
review_confidence_threshold	int	80	Minimum finding confidence (0-100) for an automated review finding to be reported.
review_gate_mode	Literal['warn', 'block']	"warn"	Whether a STANDARD or COMPREHENSIVE run that reaches deliver with no recorded review only warns (`warn`) or is blocked (`block`).
review_mandate_advisory_enabled	bool	true	Tell the agent up front that STANDARD and COMPREHENSIVE runs need a review before deliver (advisory only).
review_signoff_ttl_hours	int	24	Max validity window of an operator review sign-off, in hours.
review_verdict_ttl_hours	int	24	Hours a recorded review verdict stays valid at deliver time; older ones are treated as expired even if the reviewed bytes did not move.
risk_scaling_enabled	bool	true	Scale PRD validation thresholds by the PRD's risk level.
run_auto_close_enabled	bool	true	Close stale runs automatically when run status is assembled.
run_stale_ttl_hours	int	48	Hours without activity after which a run counts as stale and may be auto-closed.
run_staleness_grace_hours	int	12	Grace window (hours) appended to run_staleness_hours; runs inside this window emit run_near_stale_warning but are not abandoned.
run_staleness_hours	int	48	Age (hours) beyond which an active run with no activity is eligible for the stale-run sweep.
runs_root	str	".trw/runs"	Project-relative directory where run directories are created (default `.trw/runs`).
scoring_default_days_unused	int	30	Days-unused value assumed for a learning with no recorded last use.
scripts_dir	str	"scripts"	Directory under the TRW state dir that holds generated helper scripts (default `scripts`).
security	SecurityConfig	computed	Nested security settings for the MCP trust boundary.
semantic_checks_enabled	bool	true	Run semantic checks in the build gate in addition to syntax checks.
session_start_recall_enabled	bool \| None	null	Recall learnings at session start; unset follows `learning_recall_enabled`.
shared_mcp	SharedMcpConfig	computed	Opt-in shared MCP server settings: `enabled` makes clients launch `trw-mcp-proxy` against one shared server instead of one stdio server each.
skills_enabled	bool \| None	null	Install bundled skills; unset follows the client profile.
source_human_utility_boost	float	0.1	Utility bonus for learnings a human wrote or confirmed.
state_learning_default_ttl_days	dict[str, int]	computed	Default validity window in days, per learning type, proposed for a state-asserting learning (PRD-CORE-244 FR05).
sub_claude_md_max_lines	int	50	Line limit for a module-level (sub-scope) instruction file.
surface_role	Literal['agent', 'reviewer']	"agent"	MCP surface role: `agent` is the normal full surface, `reviewer` a read-only report surface for dispatched review sessions (normally set by the environment, not here).
sync_health_failure_threshold	int	10	Consecutive push failures before sync_health marks degraded.
sync_health_stale_hours	float	6.0	Hours since last successful push before sync_health marks degraded.
sync_interval_seconds	int	300	Seconds between team-sync push/pull cycles.
sync_pull_timeout_seconds	float	5.0	Seconds to wait for a pull request before giving up.
sync_push_batch_size	int	100	Most learnings pushed to the backend in one batch.
sync_push_timeout_seconds	float	10.0	Seconds to wait for a push request before giving up.
target_platforms	list[str]	computed	Platforms to sync instruction files for.
task_root	str	"docs"	Project-relative directory that holds task directories (default `docs`).
team_sync_all_projects	bool	false	Rank other projects' team learnings like unrecorded ones in recall instead of lower (default: lower).
team_sync_enabled	bool	false	Enable team sync of learnings with the backend; off by default.
team_sync_fresh_after_seconds	int	300	Pull team learnings before recall when the last pull is older than this (0: never).
telemetry	bool	false	Legacy usage-telemetry switch copied into a new project's `config.yaml` at init; the live switch is `telemetry_enabled`.
telemetry_enabled	bool	true	Record local tool-call telemetry under the logs directory.
telemetry_file	str	"tool-telemetry.jsonl"	File name of the local tool-call telemetry log (default `tool-telemetry.jsonl`).
telemetry_log_max_bytes	int	10485760	Pipeline telemetry-log rotation threshold in bytes (FR04)
templates_dir	str	"templates"	Directory under the TRW state dir that holds generated templates (default `templates`).
time_tracking_enabled	bool	true	Track elapsed time and ETAs for task-scaled runs and show them in `trw_status`, checkpoints and deliver.
timebox_hours	int	8	Default time box in hours written into a new project's `config.yaml` at init (default 8).
tool_output_size_warning_bytes	int	8192	Rendered PostToolUse tool_response byte length above which a size advisory fires (PRD-INFRA-194-FR04).
tool_resolution_mode	Literal['standard', 'all']	"standard"	Which tool packs the MCP server exposes: `standard` is the kernel plus every pack whose flag is on, `all` also enables the comms and assess packs.
traceability_coverage_min	float	0.9	Lowest allowed share of requirements traced to code or tests (0-1).
transition_nudges_enabled	bool	true	Allow phase-transition nudge lines (learning and suggestion lines) to reach the agent; independent of `nudge_enabled`.
trust_crawl_boundary	int	50	Session count up to which an agent stays in the `crawl` trust tier; above it the tier is `walk`.
trust_walk_boundary	int	200	Session count up to which an agent stays in the `walk` trust tier; above it the tier is `run`.
trw_dir	str	".trw"	Name of the TRW state directory inside the project (default `.trw`).
validation_density_weight	float	20.0	Points a PRD can earn for content density in the validation score.
validation_draft_threshold	float	60.0	Validation score at or above which a PRD is at least REVIEW quality.
validation_implementation_readiness_weight	float	25.0	Points a PRD can earn for implementation readiness in the validation score.
validation_review_threshold	float	85.0	Validation score at or above which a PRD is APPROVED quality.
validation_skeleton_threshold	float	30.0	Validation score at or above which a PRD is at least DRAFT quality (below it: skeleton).
validation_structure_weight	float	20.0	Points a PRD can earn for structure in the validation score.
validation_traceability_weight	float	35.0	Points a PRD can earn for traceability in the validation score.
verification_cache_ttl_seconds	int	3600	Freshness window for last-known evidence; never proves current-tree validity or triggers verification.
version_check_interval_seconds	float	300.0	Interval (seconds) between cached booted-vs-installed trw-mcp version-drift checks on the tool-call middleware path (PRD-CORE-215-FR02).
wal_checkpoint_max_age_seconds	int	3600	Seconds since the last SUCCESSFUL checkpoint after which one is due regardless of WAL size.
wal_checkpoint_threshold_mb	int	10	WAL size (MiB) at or above which a checkpoint is due.
wiring_gate_mode	Literal['warn', 'block']	"warn"	Whether a PRD whose requirements are not wired to code only warns (`warn`) or fails validation (`block`).
wiring_gate_mode_overrides	dict[str, Literal['warn', 'block']]	computed	Per-PRD override of `wiring_gate_mode` (PRD id or category to `warn`/`block`); empty means every PRD uses the global mode.

Retired keys

Keys that earlier releases read and this one does not. A retired key left in .trw/config.yaml is ignored, and trw-mcp warns about it and names its replacement when there is one.

Show all 212 retired keys

Retired key: adaptation_auto_approve_threshold
Replacement: None: the setting was removed

Retired key: agents_md_learning_injection
Replacement: None: the setting was removed

Retired key: agents_md_learning_max
Replacement: None: the setting was removed

Retired key: agents_md_learning_min_impact
Replacement: None: the setting was removed

Retired key: anchor_shared_set_migration_threshold
Replacement: None: the setting was removed

Retired key: api_fuzz_base_url
Replacement: None: the setting was removed

Retired key: api_fuzz_level
Replacement: None: the setting was removed

Retired key: api_fuzz_timeout_secs
Replacement: None: the setting was removed

Retired key: atdd_enabled
Replacement: None: the setting was removed

Retired key: audit_pattern_promotion_threshold
Replacement: None: the setting was removed

Retired key: auto_checkpoint_enabled
Replacement: None: the setting was removed

Retired key: auto_checkpoint_tool_interval
Replacement: None: the setting was removed

Retired key: auto_upgrade
Replacement: None: the setting was removed

Retired key: build_check_mypy_args
Replacement: None: the setting was removed

Retired key: build_check_pytest_args
Replacement: None: the setting was removed

Retired key: build_check_pytest_cmd
Replacement: None: the setting was removed

Retired key: build_check_timeout_secs
Replacement: None: the setting was removed

Retired key: build_freshness_window_secs
Replacement: None: the setting was removed

Retired key: ceremony_alert_consecutive
Replacement: None: the setting was removed

Retired key: ceremony_alert_threshold
Replacement: None: the setting was removed

Retired key: checkpoint_secs
Replacement: None: the setting was removed

Retired key: checkpoints_file
Replacement: None: the setting was removed

Retired key: claude_md_max_lines
Replacement: None: the setting was removed

Retired key: code_index_enabled
Replacement: None: the setting was removed

Retired key: commit_fr_trailer_enabled
Replacement: None: the setting was removed

Retired key: comms_group_admission_limit
Replacement: comms_group_row_limit

Retired key: compact_after_turns
Replacement: None: the setting was removed

Retired key: completion_hooks_blocking
Replacement: None: the setting was removed

Retired key: compliance_history_file
Replacement: None: the setting was removed

Retired key: compliance_long_session_event_threshold
Replacement: None: the setting was removed

Retired key: compliance_pass_threshold
Replacement: None: the setting was removed

Retired key: compliance_strictness
Replacement: None: the setting was removed

Retired key: compliance_warning_threshold
Replacement: None: the setting was removed

Retired key: confidence_threshold
Replacement: None: the setting was removed

Retired key: consensus_quorum
Replacement: None: the setting was removed

Retired key: consistency_validation_min
Replacement: None: the setting was removed

Retired key: contradiction_penalty_reward
Replacement: None: the setting was removed

Retired key: debt_actionable_threshold
Replacement: None: the setting was removed

Retired key: debt_auto_promote_threshold
Replacement: None: the setting was removed

Retired key: debt_budget_critical_ratio
Replacement: None: the setting was removed

Retired key: debt_budget_high_ratio
Replacement: None: the setting was removed

Retired key: debt_decay_assessment_rate
Replacement: None: the setting was removed

Retired key: debt_decay_base_score
Replacement: None: the setting was removed

Retired key: debt_decay_daily_rate
Replacement: None: the setting was removed

Retired key: debt_default_wave_size
Replacement: None: the setting was removed

Retired key: debt_id_prefix
Replacement: None: the setting was removed

Retired key: debt_initial_decay_score
Replacement: None: the setting was removed

Retired key: debt_registry_filename
Replacement: None: the setting was removed

Retired key: decision_enabled
Replacement: assess_enabled

Retired key: density_weight_default
Replacement: None: the setting was removed

Retired key: density_weight_functional_requirements
Replacement: None: the setting was removed

Retired key: density_weight_problem_statement
Replacement: None: the setting was removed

Retired key: density_weight_traceability_matrix
Replacement: None: the setting was removed

Retired key: dispatch_role_client
Replacement: dispatch_default_client (or pass --client); a role no longer picks the client

Retired key: doctor_thread_hotspot_min_seconds
Replacement: None: the setting was removed

Retired key: doctor_thread_hotspot_share
Replacement: None: the setting was removed

Retired key: embeddings_auto_backfill_on_low_coverage
Replacement: None: the setting was removed

Retired key: external_store_recall_cap
Replacement: None: the setting was removed

Retired key: extra_read_stores
Replacement: None: the setting was removed

Retired key: finding_dedup_threshold
Replacement: None: the setting was removed

Retired key: findings_dir
Replacement: None: the setting was removed

Retired key: findings_entries_dir
Replacement: None: the setting was removed

Retired key: findings_registry_file
Replacement: None: the setting was removed

Retired key: framework_overhead_threshold
Replacement: None: the setting was removed

Retired key: gate_architecture_score_penalty
Replacement: None: the setting was removed

Retired key: gate_critic_overhead_multiplier
Replacement: None: the setting was removed

Retired key: gate_tokens_per_1k_chars
Replacement: None: the setting was removed

Retired key: grooming_max_iterations
Replacement: None: the setting was removed

Retired key: grooming_partial_density_threshold
Replacement: None: the setting was removed

Retired key: grooming_placeholder_density_threshold
Replacement: None: the setting was removed

Retired key: grooming_research_scope
Replacement: None: the setting was removed

Retired key: grooming_target_completeness
Replacement: None: the setting was removed

Retired key: hint_sidecar_auto_refresh_enabled
Replacement: hint_sidecar_refresh_enabled (the pre-edit hint no longer requests builds; post-commit and session start do)

Retired key: hint_sidecar_refresh_file_cap
Replacement: None: the setting was removed

Retired key: hybrid_bm25_candidates
Replacement: None: the setting was removed

Retired key: hybrid_search_candidate_pool_size
Replacement: None: the setting was removed

Retired key: hybrid_vector_candidates
Replacement: None: the setting was removed

Retired key: impact_decay_half_life_days
Replacement: None: the setting was removed

Retired key: impact_forced_distribution_enabled
Replacement: None: the setting was removed

Retired key: impact_high_threshold_pct
Replacement: None: the setting was removed

Retired key: impact_tier_critical_cap
Replacement: None: the setting was removed

Retired key: impact_tier_high_cap
Replacement: None: the setting was removed

Retired key: incremental_validation_enabled
Replacement: None: the setting was removed

Retired key: instruction_external_filename
Replacement: None: the setting was removed

Retired key: instruction_externalize
Replacement: None: the setting was removed

Retired key: learn_journal_drain_min_batch
Replacement: None: the setting was removed

Retired key: learn_journal_pending_max_age_hours
Replacement: None: the setting was removed

Retired key: learning_outcome_history_cap
Replacement: None: the setting was removed

Retired key: learning_promotion_impact
Replacement: None: the setting was removed

Retired key: learning_prune_age_days
Replacement: None: the setting was removed

Retired key: lifecycle_use_fsrs
Replacement: None: the setting was removed

Retired key: llm_utility_filter_enabled
Replacement: None: the setting was removed

Retired key: max_audit_cycles
Replacement: None: the setting was removed

Retired key: max_child_depth
Replacement: None: the setting was removed

Retired key: max_cluster_size
Replacement: None: the setting was removed

Retired key: max_research_waves
Replacement: None: the setting was removed

Retired key: mcp_server_instructions_enabled
Replacement: None: the setting was removed

Retired key: memory_cold_threshold_days
Replacement: None: the setting was removed

Retired key: memory_consolidation_interval_days
Replacement: None: the setting was removed

Retired key: memory_decay_batch_size
Replacement: MEMORY_DECAY_BATCH_SIZE (environment variable, daemon-side MemoryConfig.decay_batch_size)

Retired key: memory_decay_cutoff_days
Replacement: MEMORY_DECAY_CUTOFF_DAYS (environment variable, daemon-side MemoryConfig.decay_cutoff_days)

Retired key: memory_hot_ttl_days
Replacement: None: the setting was removed

Retired key: memory_retention_days
Replacement: None: the setting was removed

Retired key: memory_store_path
Replacement: None: the setting was removed

Retired key: min_shards_floor
Replacement: None: the setting was removed

Retired key: min_shards_target
Replacement: None: the setting was removed

Retired key: minimal_after_turns
Replacement: None: the setting was removed

Retired key: mutation_critical_paths
Replacement: None: the setting was removed

Retired key: mutation_enabled
Replacement: None: the setting was removed

Retired key: mutation_experimental_paths
Replacement: None: the setting was removed

Retired key: mutation_threshold
Replacement: None: the setting was removed

Retired key: mutation_threshold_critical
Replacement: None: the setting was removed

Retired key: mutation_threshold_experimental
Replacement: None: the setting was removed

Retired key: mutation_timeout_secs
Replacement: None: the setting was removed

Retired key: nudge_budget_chars
Replacement: None: the setting was removed

Retired key: nudge_dedup_enabled
Replacement: None: the setting was removed

Retired key: nudge_pool_weight_ceremony
Replacement: nudge_pool_weights.ceremony

Retired key: nudge_pool_weight_context
Replacement: nudge_pool_weights.context

Retired key: nudge_pool_weight_learnings
Replacement: nudge_pool_weights.learnings

Retired key: nudge_pool_weight_workflow
Replacement: nudge_pool_weights.workflow

Retired key: nudge_urgency_mode
Replacement: None: the setting was removed

Retired key: observation_masking
Replacement: None: the setting was removed

Retired key: otel_capture_messages
Replacement: None: the setting was removed

Retired key: otel_endpoint
Replacement: None: the setting was removed

Retired key: otel_semconv
Replacement: None: the setting was removed

Retired key: outcome_weight_learning_rate
Replacement: None: the setting was removed

Retired key: outcome_weight_p0_defects
Replacement: None: the setting was removed

Retired key: outcome_weight_rework
Replacement: None: the setting was removed

Retired key: outcome_weight_velocity
Replacement: None: the setting was removed

Retired key: pause_after_compaction
Replacement: None: the setting was removed

Retired key: phase_cap_deliver
Replacement: None: the setting was removed

Retired key: phase_cap_implement
Replacement: None: the setting was removed

Retired key: phase_cap_plan
Replacement: None: the setting was removed

Retired key: phase_cap_research
Replacement: None: the setting was removed

Retired key: phase_cap_review
Replacement: None: the setting was removed

Retired key: phase_cap_validate
Replacement: None: the setting was removed

Retired key: phase_exposure_enabled
Replacement: None: the setting was removed

Retired key: phase_transition_withhold_rate
Replacement: None: the setting was removed

Retired key: pipeline_health_bandit_probe_enabled
Replacement: None: the setting was removed

Retired key: pipeline_health_bandit_stale_days
Replacement: None: the setting was removed

Retired key: provenance_enabled
Replacement: None: the setting was removed

Retired key: proximal_reward_weight
Replacement: None: the setting was removed

Retired key: q_cold_start_threshold
Replacement: None: the setting was removed

Retired key: q_learning_rate
Replacement: None: the setting was removed

Retired key: q_recurrence_bonus
Replacement: None: the setting was removed

Retired key: recall_receipt_max_entries
Replacement: None: the setting was removed

Retired key: recall_verification_budget_ms
Replacement: None: the setting was removed

Retired key: reflect_max_positive_learnings
Replacement: None: the setting was removed

Retired key: reflect_q_value_threshold
Replacement: None: the setting was removed

Retired key: reflect_sequence_lookback
Replacement: None: the setting was removed

Retired key: retrieval_embedding_dim
Replacement: None: the setting was removed

Retired key: retrieval_embedding_model
Replacement: MEMORY_EMBEDDING_MODEL (environment variable, daemon-wide)

Retired key: reversion_rate_concerning
Replacement: None: the setting was removed

Retired key: reversion_rate_elevated
Replacement: None: the setting was removed

Retired key: run_archive_hours
Replacement: None: the setting was removed

Retired key: run_auto_close_age_days
Replacement: run_stale_ttl_hours

Retired key: scoring_error_fallback_reward
Replacement: None: the setting was removed

Retired key: scoring_error_keywords
Replacement: None: the setting was removed

Retired key: scoring_recency_discount_floor
Replacement: None: the setting was removed

Retired key: scout_blast_radius_threshold
Replacement: None: the setting was removed

Retired key: scout_churn_commit_threshold
Replacement: None: the setting was removed

Retired key: scout_enabled
Replacement: None: the setting was removed

Retired key: scout_max_mode3_rate
Replacement: None: the setting was removed

Retired key: scratch_dir
Replacement: None: the setting was removed

Retired key: security_check_enabled
Replacement: None: the setting was removed

Retired key: self_review_blocking
Replacement: None: the setting was removed

Retired key: session_start_defer_under_writer_pressure
Replacement: None: the setting was removed

Retired key: session_start_max_deferral_hours
Replacement: None: the setting was removed

Retired key: session_start_recent_bypass_days
Replacement: None: the setting was removed

Retired key: session_start_recent_bypass_min_impact
Replacement: None: the setting was removed

Retired key: session_start_writer_pressure_threshold
Replacement: None: the setting was removed

Retired key: simplifier_backup_dir
Replacement: None: the setting was removed

Retired key: simplifier_verification_timeout_secs
Replacement: None: the setting was removed

Retired key: simplifier_wave_size
Replacement: None: the setting was removed

Retired key: skill_active_cap
Replacement: None: the setting was removed

Retired key: skill_contribution_cold_start
Replacement: None: the setting was removed

Retired key: skill_contribution_half_life_days
Replacement: None: the setting was removed

Retired key: skill_duplicate_max_skills
Replacement: None: the setting was removed

Retired key: skill_duplicate_similarity_threshold
Replacement: None: the setting was removed

Retired key: skill_retirement_floor
Replacement: None: the setting was removed

Retired key: skill_retirement_windows
Replacement: None: the setting was removed

Retired key: skill_surface_tracking_enabled
Replacement: None: the setting was removed

Retired key: source_package_name
Replacement: None: the setting was removed

Retired key: source_package_path
Replacement: None: the setting was removed

Retired key: sprint_code_simplifier_wave_size
Replacement: None: the setting was removed

Retired key: sprint_commit_pattern
Replacement: None: the setting was removed

Retired key: sprint_integration_branch_pattern
Replacement: None: the setting was removed

Retired key: strict_input_criteria
Replacement: None: the setting was removed

Retired key: test_map_filename
Replacement: None: the setting was removed

Retired key: test_skeleton_dir
Replacement: None: the setting was removed

Retired key: tests_relative_path
Replacement: None: the setting was removed

Retired key: tool_access_grant_max_ttl_seconds
Replacement: None: the setting was removed

Retired key: tool_descriptions_variant
Replacement: None: the setting was removed

Retired key: trust_locked
Replacement: None: the setting was removed

Retired key: trust_security_tags
Replacement: None: the setting was removed

Retired key: trust_walk_sample_rate
Replacement: None: the setting was removed

Retired key: update_channel
Replacement: None: the setting was removed

Retired key: user_tier_enabled
Replacement: None: the setting was removed

Retired key: validation_ears_weight
Replacement: None: the setting was removed

Retired key: validation_fk_optimal_max
Replacement: None: the setting was removed

Retired key: validation_fk_optimal_min
Replacement: None: the setting was removed

Retired key: validation_readability_weight
Replacement: None: the setting was removed

Retired key: validation_smell_weight
Replacement: None: the setting was removed

Retired key: velocity_alert_min_runs
Replacement: None: the setting was removed

Retired key: velocity_alert_r_squared_min
Replacement: None: the setting was removed

Retired key: velocity_confounder_jump_ratio
Replacement: None: the setting was removed

Retired key: velocity_effective_q_threshold
Replacement: None: the setting was removed

Retired key: velocity_history_max_entries
Replacement: None: the setting was removed

Retired key: velocity_sign_test_alpha
Replacement: None: the setting was removed

Retired key: velocity_stable_threshold
Replacement: None: the setting was removed

Retired key: wal_checkpoint_idle_interval_seconds
Replacement: None: the setting was removed

Retired key: worktree_dir
Replacement: None: the setting was removed

Retired key	Replacement
adaptation_auto_approve_threshold	None: the setting was removed
agents_md_learning_injection	None: the setting was removed
agents_md_learning_max	None: the setting was removed
agents_md_learning_min_impact	None: the setting was removed
anchor_shared_set_migration_threshold	None: the setting was removed
api_fuzz_base_url	None: the setting was removed
api_fuzz_level	None: the setting was removed
api_fuzz_timeout_secs	None: the setting was removed
atdd_enabled	None: the setting was removed
audit_pattern_promotion_threshold	None: the setting was removed
auto_checkpoint_enabled	None: the setting was removed
auto_checkpoint_tool_interval	None: the setting was removed
auto_upgrade	None: the setting was removed
build_check_mypy_args	None: the setting was removed
build_check_pytest_args	None: the setting was removed
build_check_pytest_cmd	None: the setting was removed
build_check_timeout_secs	None: the setting was removed
build_freshness_window_secs	None: the setting was removed
ceremony_alert_consecutive	None: the setting was removed
ceremony_alert_threshold	None: the setting was removed
checkpoint_secs	None: the setting was removed
checkpoints_file	None: the setting was removed
claude_md_max_lines	None: the setting was removed
code_index_enabled	None: the setting was removed
commit_fr_trailer_enabled	None: the setting was removed
comms_group_admission_limit	comms_group_row_limit
compact_after_turns	None: the setting was removed
completion_hooks_blocking	None: the setting was removed
compliance_history_file	None: the setting was removed
compliance_long_session_event_threshold	None: the setting was removed
compliance_pass_threshold	None: the setting was removed
compliance_strictness	None: the setting was removed
compliance_warning_threshold	None: the setting was removed
confidence_threshold	None: the setting was removed
consensus_quorum	None: the setting was removed
consistency_validation_min	None: the setting was removed
contradiction_penalty_reward	None: the setting was removed
debt_actionable_threshold	None: the setting was removed
debt_auto_promote_threshold	None: the setting was removed
debt_budget_critical_ratio	None: the setting was removed
debt_budget_high_ratio	None: the setting was removed
debt_decay_assessment_rate	None: the setting was removed
debt_decay_base_score	None: the setting was removed
debt_decay_daily_rate	None: the setting was removed
debt_default_wave_size	None: the setting was removed
debt_id_prefix	None: the setting was removed
debt_initial_decay_score	None: the setting was removed
debt_registry_filename	None: the setting was removed
decision_enabled	assess_enabled
density_weight_default	None: the setting was removed
density_weight_functional_requirements	None: the setting was removed
density_weight_problem_statement	None: the setting was removed
density_weight_traceability_matrix	None: the setting was removed
dispatch_role_client	dispatch_default_client (or pass --client); a role no longer picks the client
doctor_thread_hotspot_min_seconds	None: the setting was removed
doctor_thread_hotspot_share	None: the setting was removed
embeddings_auto_backfill_on_low_coverage	None: the setting was removed
external_store_recall_cap	None: the setting was removed
extra_read_stores	None: the setting was removed
finding_dedup_threshold	None: the setting was removed
findings_dir	None: the setting was removed
findings_entries_dir	None: the setting was removed
findings_registry_file	None: the setting was removed
framework_overhead_threshold	None: the setting was removed
gate_architecture_score_penalty	None: the setting was removed
gate_critic_overhead_multiplier	None: the setting was removed
gate_tokens_per_1k_chars	None: the setting was removed
grooming_max_iterations	None: the setting was removed
grooming_partial_density_threshold	None: the setting was removed
grooming_placeholder_density_threshold	None: the setting was removed
grooming_research_scope	None: the setting was removed
grooming_target_completeness	None: the setting was removed
hint_sidecar_auto_refresh_enabled	hint_sidecar_refresh_enabled (the pre-edit hint no longer requests builds; post-commit and session start do)
hint_sidecar_refresh_file_cap	None: the setting was removed
hybrid_bm25_candidates	None: the setting was removed
hybrid_search_candidate_pool_size	None: the setting was removed
hybrid_vector_candidates	None: the setting was removed
impact_decay_half_life_days	None: the setting was removed
impact_forced_distribution_enabled	None: the setting was removed
impact_high_threshold_pct	None: the setting was removed
impact_tier_critical_cap	None: the setting was removed
impact_tier_high_cap	None: the setting was removed
incremental_validation_enabled	None: the setting was removed
instruction_external_filename	None: the setting was removed
instruction_externalize	None: the setting was removed
learn_journal_drain_min_batch	None: the setting was removed
learn_journal_pending_max_age_hours	None: the setting was removed
learning_outcome_history_cap	None: the setting was removed
learning_promotion_impact	None: the setting was removed
learning_prune_age_days	None: the setting was removed
lifecycle_use_fsrs	None: the setting was removed
llm_utility_filter_enabled	None: the setting was removed
max_audit_cycles	None: the setting was removed
max_child_depth	None: the setting was removed
max_cluster_size	None: the setting was removed
max_research_waves	None: the setting was removed
mcp_server_instructions_enabled	None: the setting was removed
memory_cold_threshold_days	None: the setting was removed
memory_consolidation_interval_days	None: the setting was removed
memory_decay_batch_size	MEMORY_DECAY_BATCH_SIZE (environment variable, daemon-side MemoryConfig.decay_batch_size)
memory_decay_cutoff_days	MEMORY_DECAY_CUTOFF_DAYS (environment variable, daemon-side MemoryConfig.decay_cutoff_days)
memory_hot_ttl_days	None: the setting was removed
memory_retention_days	None: the setting was removed
memory_store_path	None: the setting was removed
min_shards_floor	None: the setting was removed
min_shards_target	None: the setting was removed
minimal_after_turns	None: the setting was removed
mutation_critical_paths	None: the setting was removed
mutation_enabled	None: the setting was removed
mutation_experimental_paths	None: the setting was removed
mutation_threshold	None: the setting was removed
mutation_threshold_critical	None: the setting was removed
mutation_threshold_experimental	None: the setting was removed
mutation_timeout_secs	None: the setting was removed
nudge_budget_chars	None: the setting was removed
nudge_dedup_enabled	None: the setting was removed
nudge_pool_weight_ceremony	nudge_pool_weights.ceremony
nudge_pool_weight_context	nudge_pool_weights.context
nudge_pool_weight_learnings	nudge_pool_weights.learnings
nudge_pool_weight_workflow	nudge_pool_weights.workflow
nudge_urgency_mode	None: the setting was removed
observation_masking	None: the setting was removed
otel_capture_messages	None: the setting was removed
otel_endpoint	None: the setting was removed
otel_semconv	None: the setting was removed
outcome_weight_learning_rate	None: the setting was removed
outcome_weight_p0_defects	None: the setting was removed
outcome_weight_rework	None: the setting was removed
outcome_weight_velocity	None: the setting was removed
pause_after_compaction	None: the setting was removed
phase_cap_deliver	None: the setting was removed
phase_cap_implement	None: the setting was removed
phase_cap_plan	None: the setting was removed
phase_cap_research	None: the setting was removed
phase_cap_review	None: the setting was removed
phase_cap_validate	None: the setting was removed
phase_exposure_enabled	None: the setting was removed
phase_transition_withhold_rate	None: the setting was removed
pipeline_health_bandit_probe_enabled	None: the setting was removed
pipeline_health_bandit_stale_days	None: the setting was removed
provenance_enabled	None: the setting was removed
proximal_reward_weight	None: the setting was removed
q_cold_start_threshold	None: the setting was removed
q_learning_rate	None: the setting was removed
q_recurrence_bonus	None: the setting was removed
recall_receipt_max_entries	None: the setting was removed
recall_verification_budget_ms	None: the setting was removed
reflect_max_positive_learnings	None: the setting was removed
reflect_q_value_threshold	None: the setting was removed
reflect_sequence_lookback	None: the setting was removed
retrieval_embedding_dim	None: the setting was removed
retrieval_embedding_model	MEMORY_EMBEDDING_MODEL (environment variable, daemon-wide)
reversion_rate_concerning	None: the setting was removed
reversion_rate_elevated	None: the setting was removed
run_archive_hours	None: the setting was removed
run_auto_close_age_days	run_stale_ttl_hours
scoring_error_fallback_reward	None: the setting was removed
scoring_error_keywords	None: the setting was removed
scoring_recency_discount_floor	None: the setting was removed
scout_blast_radius_threshold	None: the setting was removed
scout_churn_commit_threshold	None: the setting was removed
scout_enabled	None: the setting was removed
scout_max_mode3_rate	None: the setting was removed
scratch_dir	None: the setting was removed
security_check_enabled	None: the setting was removed
self_review_blocking	None: the setting was removed
session_start_defer_under_writer_pressure	None: the setting was removed
session_start_max_deferral_hours	None: the setting was removed
session_start_recent_bypass_days	None: the setting was removed
session_start_recent_bypass_min_impact	None: the setting was removed
session_start_writer_pressure_threshold	None: the setting was removed
simplifier_backup_dir	None: the setting was removed
simplifier_verification_timeout_secs	None: the setting was removed
simplifier_wave_size	None: the setting was removed
skill_active_cap	None: the setting was removed
skill_contribution_cold_start	None: the setting was removed
skill_contribution_half_life_days	None: the setting was removed
skill_duplicate_max_skills	None: the setting was removed
skill_duplicate_similarity_threshold	None: the setting was removed
skill_retirement_floor	None: the setting was removed
skill_retirement_windows	None: the setting was removed
skill_surface_tracking_enabled	None: the setting was removed
source_package_name	None: the setting was removed
source_package_path	None: the setting was removed
sprint_code_simplifier_wave_size	None: the setting was removed
sprint_commit_pattern	None: the setting was removed
sprint_integration_branch_pattern	None: the setting was removed
strict_input_criteria	None: the setting was removed
test_map_filename	None: the setting was removed
test_skeleton_dir	None: the setting was removed
tests_relative_path	None: the setting was removed
tool_access_grant_max_ttl_seconds	None: the setting was removed
tool_descriptions_variant	None: the setting was removed
trust_locked	None: the setting was removed
trust_security_tags	None: the setting was removed
trust_walk_sample_rate	None: the setting was removed
update_channel	None: the setting was removed
user_tier_enabled	None: the setting was removed
validation_ears_weight	None: the setting was removed
validation_fk_optimal_max	None: the setting was removed
validation_fk_optimal_min	None: the setting was removed
validation_readability_weight	None: the setting was removed
validation_smell_weight	None: the setting was removed
velocity_alert_min_runs	None: the setting was removed
velocity_alert_r_squared_min	None: the setting was removed
velocity_confounder_jump_ratio	None: the setting was removed
velocity_effective_q_threshold	None: the setting was removed
velocity_history_max_entries	None: the setting was removed
velocity_sign_test_alpha	None: the setting was removed
velocity_stable_threshold	None: the setting was removed
wal_checkpoint_idle_interval_seconds	None: the setting was removed
worktree_dir	None: the setting was removed