Layer 1 of 6
Requirements Engineering
AI stops building the wrong thing
Turns vague requests into structured specs with acceptance criteria — agents verify work against the spec, not vibes.
See requirements engineering in motionEngineering operating layerEngineering operating layer for AI agents
Your AI can write code. TRW makes it shippable.
TRW sits above your model, IDE, and memory. It keeps requirements, execution, verification, handoff, and preserved project knowledge connected from one session to the next.
AI Agent — TRW session
$
Knowledge base
4,369learnings · from this repo
Database migrations must run before deployment, not after
deploymentdeploy-pipeline
critical
0.95
JWT refresh tokens need rotation on every use
securityauth-service
critical
0.93
React useEffect cleanup prevents memory leaks on unmount
frontendplatform-ui
high
0.85
Representative entries · local-first
Install the engineering operating layer your AI tools are missing
Open public beta · Free framework · Create a free account to install
Designed for the AI coding ecosystem
Claude Code
GitHub Copilot
Cursor
OpenCode
Codex
LangChain
LlamaIndex
CrewAI
OpenAI
Antigravity
From one session to the next
The next session starts with the project's receipts.
TRW does not promise that an agent gets smarter after every run. It preserves the requirements, decisions, validation results, and gotchas that the next run would otherwise have to rediscover.
Without a handoff
Context resets. Work repeats.
- 01Re-read the repository
- 02Re-ask the constraints
- 03Re-find the same gotcha
The model can still write code. Your team pays again for the context it already earned.
With a TRW handoff
Session 48 leaves a trail.
decisionKeep authentication server-side
verificationVisual route matrix passed
gotchaRun migrations before deploy
Session 49 receives only what fits the task
Relevant records, source-linked and ready to inspect.
The advantage is concrete: less rediscovery, traceable decisions, and evidence that remains readable after the chat is gone.
Inspect the memory systemGoverned execution
A workflow that keeps specs, code, and verification connected
Research, plan, implement, validate, review, and deliver — with evidence at each handoff and reusable knowledge at the end.
Research
Plan
Implement
Validate
Review
Deliver
Research
Recall prior work & inspect the codebase
Plan
Choose the approach & map dependencies
Implement
Ship with checkpoints so work survives compaction
Validate
Run tests, type checks & coverage gates
Review
Audit correctness, quality & security
Deliver
Publish artifacts & persist learnings
Run complete · learnings signed
All phases signed
What's Inside
Six capabilities that turn AI output into engineering work
Memory is one piece of the system. Requirements, verification, handoffs, and coordination are what make the work trustworthy.
Evidence ledger
The numbers. Their source. The limits of the claim.
TRW runs on its own codebase. We publish the operational evidence it produces—and keep that separate from outcomes we have not proven yet.
Patterns Discovered
4,000
Gotchas, patterns, and decisions your agents never have to rediscover.
Agent Actions Tracked
1,000
Tool calls, checkpoints, and deliveries captured across TRW runs in this repo.
Projects Using TRW
1
Projects reporting TRW installs via opt-in telemetry. When telemetry is unavailable, we show this repo’s own numbers — never invented ones.
Quality Compliance
—
How consistently agents follow the engineering workflow before shipping.
Read the evidence at the right confidence level
Product telemetry is useful. It is not a substitute for an outcome study.
Evidence level 01
Confirmed in this repo
Requirements, checkpoints, build results, reviews, and delivery records are persisted as inspectable artifacts.
Evidence level 02
Observed and counted
TRW records recalled learnings and workflow events. The figures above come from live telemetry when available, otherwise a labeled repository snapshot.
Evidence level 03
Still an open question
Whether retained context improves task success across teams, models, and repositories requires controlled evaluation. We do not turn usage counts into that claim.
Open public beta · Free framework · No approval required
What survives the handoff
A completed task leaves more than a diff.
TRW packages the reasoning around the code so the next engineer or agent can verify the result without reconstructing the entire conversation.
- 01
Intent stays attached
The next reader can see why the change exists.
- 02
Evidence stays verifiable
Reviewers can inspect what ran instead of trusting a claim.
- 03
The next step stays explicit
A new session resumes from a known state, not a blank chat.
Signup hardening
task 24-071 · completed
Work accepted
The diff is only one output
02 linked
Requirements
Success criteria
03 saved
Decisions
With rationale
12 passed
Verification
Commands + output
04 ranked
Learnings
Ready for recall
Next session
Relevant context is ready
scopedsource-linkedinspectable
Repo-local baseline
State stays beside the work it describes.
Human-readable
Plain files remain inspectable without TRW.
Remote sync is opt-in
Local operation does not require hosted storage.
Install
Install once. Improve the next session.
TRW wires in as a single configuration entry. No app code changes. No migrations. Your existing tools stay in place. Remove it anytime — zero lock-in.
TTY_INSTALLone command
| 1 | # Install or upgrade TRW in your project |
| 2 | curl -fsSL https://trwframework.com/install.sh | bash |
Offline by default
Nothing leaves your machine. Telemetry is opt-in.
Plain-file state
Project knowledge lives in readable repo-local files that you can inspect, version, archive, or remove.
Language-agnostic
TRW works with any codebase. Verification hooks can run your existing test stack, not a new one.
FAQ
Common questions
AI writes code. TRW makes it engineering.
Start with the docs, create a free account, and keep requirements, verification, and learnings connected from one session to the next.
Open public beta · Free framework · Plain-file repo-local state