Toolkit
Toolkit
The ADLC toolkit: one deterministic gate per lifecycle phase, installed once through @adlc/cli.
Toolkit
The toolkit makes the ADLC gates concrete. Each lifecycle phase asks one
machine-checkable question, and one tool answers it with an exit code: 0
passes, 2 fails. The tools publish independently but install together and share
one command surface: adlc <tool>. If you want the phase map first, read
The Lifecycle; this page is the index of every gate.
npm install -g @adlc/cli
adlc --help # list every tool, grouped
adlc <tool> [args] # run any gateEverything below is reachable as adlc <tool>. The groups mirror the phases of
work: shaping the spec, supervising execution, gathering review evidence,
compounding defenses, and syncing outward.
Spec and ticket shaping
| Tool | What it does |
|---|---|
| tickets | Canonical sharded ticket store, migration bridge, transactions, archive, and provenance. |
| parallax | Fan out readers to expose spec ambiguity and route conflicts. |
| spec-lint | Gate specs for acceptance criteria that lack a verification method. |
| premortem | Stress-test an approved spec before implementation. |
| coldstart | Check whether tickets are executable without agent guesswork. |
| ticket-prune | Report and archive stale, already-shipped tickets out of the active store. |
Execution supervision and rails
| Tool | What it does |
|---|---|
| preflight | Check baseline environment readiness before fan-out. |
| model-router | Assign tickets to frontier, direct, or ladder model strategies. |
| merge-forecast | Estimate fan-out width, dependency pressure, and merge backpressure. |
| rails-guard | Enforce frozen rails, declared suppressions, and manifest recording. |
| build-gate | Deny starting a high-risk ticket build in a degraded (context-rot) session unless audited. |
| flail-detector | Detect repeated errors, scope violations, edit churn, and oversized logs. |
| consensus-fix | Fan out candidate fixes and select the gated consensus winner. |
Review evidence and calibration
| Tool | What it does |
|---|---|
| behavior-diff | Capture and compare HTTP/API behavior snapshots for the P6 human gate. |
| gate-manifest | Record, verify, show, and attest append-only gate evidence. |
| hollow-test | Mutate changed code to find tests that pass without testing behavior. |
| prosecute | Record ticket- and revision-bound P5 prosecution evidence. |
| review-calibration | Measure reviewer recall by scoring whether review catches mutants. |
| model-ratchet | Identify hot files for re-prosecution after model or repo drift. |
| gate-fuzzing | Run hostile candidates against gate suites to find defeats. |
review | External passthrough. Runs the model-judged npx adversarial-review that prosecute records evidence for. |
Compounding defenses
| Tool | What it does |
|---|---|
| lesson-foundry | Mine repeated findings into deterministic defenses. |
| rejection-mining | Mine review rejections into reusable review lenses. |
| skill-rot | Check skill files for stale validation metadata and stamp freshness. |
External integration
| Tool | What it does |
|---|---|
| ticket-sync | Two-way sync ADLC tickets with external trackers (GitHub Issues). Reached as adlc ticket. |
Shared foundation
These are not gates. They are the layer every gate is built on.