Repository context coverage and agent operating guidance.
dev.kit turns repo design into a usable contract for agents.
It does three things:
- inspect what the current environment can really support
- detect and serialize repo context into
.rabbit/context.yaml - generate
AGENTS.mdso each new session starts from current repo reality instead of prompt memory
The model is repo-first, gap-aware, and regeneration-friendly. dev.kit should describe what the repo declares, note what it cannot confirm yet, and make the next repair step obvious.
npm install -g @udx/dev-kit# first make sure your dev.kit install is current
# npm install -g @udx/dev-kit
# or: curl -fsSL https://raw.githubusercontent.com/udx/dev.kit/latest/bin/scripts/install.sh | bash
cd my-repo
dev.kit # happy path: env + repo context + AGENTS.md
dev.kit env # inspect tools, auth, and capability controls
dev.kit env --config
dev.kit repo # refresh only .rabbit/context.yaml
dev.kit agent # refresh only AGENTS.mdThe intended loop is simple:
- make sure the local
dev.kitinstall is current - run
dev.kitat the start of a session - let
dev.kit envshape what capabilities are actually available - let
dev.kit repowrite the current repo contract into.rabbit/context.yaml - let
dev.kit agentgenerate operating guidance from that contract - if gaps are detected, fix the repo-owned source assets, rerun
dev.kit repo, then validate the regenerated context
That keeps context dynamic, grounded in repo signals, and resistant to drift.
| Command | Role |
|---|---|
dev.kit |
Start here. Refresh environment awareness, repo context, and agent guidance together. |
dev.kit env |
Detect tools, auth state, and local capability controls so later steps stay honest. |
dev.kit env --config |
Create or update env config for disabling specific tools or credentials. |
dev.kit repo |
Detect refs, commands, gaps, manifests, and dependencies, then write .rabbit/context.yaml. |
dev.kit repo --force |
Re-resolve dependency context from scratch. |
dev.kit agent |
Generate AGENTS.md from the current repo contract and its gaps. |
All commands support --json for machine-readable output and should guide the next step in human- and agent-friendly terms.
dev.kit produces two main artifacts:
.rabbit/context.yaml— the machine-readable repo contractAGENTS.md— the generated operating layer for agents
Keep the boundary strict:
context.yamlis for repo facts, traces, commands, manifests, dependencies, and gapsAGENTS.mdis for how an agent should operate from that contract, including gap-repair behavior
# npm (recommended)
npm install -g @udx/dev-kit
# no npm?
curl -fsSL https://raw.githubusercontent.com/udx/dev.kit/latest/bin/scripts/install.sh | bashUse one install path at a time. Installing with npm removes the curl-managed home and shim. Installing with curl removes the global npm package first. More detail: Installation.
- How It Works — command flow, generated artifacts, and regeneration loop
- Environment Config — capability detection and env controls
- Context Coverage — what
context.yamlshould contain and what gaps mean - Experience Guidance — what
AGENTS.mdshould instruct agents to do - Smart Dependency Detection — deterministic cross-repo and manifest tracing
- Installation — npm and curl installs, cleanup, uninstall, and verification
For fast local checks:
bash tests/suite.sh --only coreFor installed-CLI testing in a real worker environment:
bash tests/worker-smoke.shFor opt-in validation against real local repos:
bash tests/real-repos.sh /path/to/repo1 /path/to/repo2The worker runner is the main integration path for heavier scenarios such as gap repair, env toggles, and real-repo mutation. Real-repo testing is local-only and can include both public and private repos without baking those assumptions into CI.