Menรผ
Route software work through the Agentic Development Lifecycle in Cursor. Use when the user asks to apply ADLC, operate in ADLC, triage work, choose the next ADLC gate, or mentions P0โP7 / rails / prosecute.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fรผgen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prรผfen und installieren.
Basierend auf der SOC-Berufsklassifikation
Refute-charter review discipline for prosecuting a diff against its ticket. Use when asked to act as a prosecutor, reviewer, or to find reasons a change must not merge.
Run ADLC P7 distillation and maintenance workflows in Codex, including lesson-foundry, rejection-mining, skill-mining, scheduled maintenance, skill-rot, model-ratchet, review-calibration, and gate-fuzzing.
Run the multi-lens P5 adversarial prosecution fan-out and record ADLC P5 review evidence and P6 acceptance packet workflows in Codex. Fans out five independent read-only lenses, dedupes findings, verifies each independently, and loops until dry before recording evidence.
Run ADLC P7 distillation in the GitHub Copilot CLI โ lesson-foundry, rejection-mining, and review-calibration โ turning repeated review findings and PR rejections into permanent, deterministic defenses. Decay-driven maintenance lives in the adlc-maintain skill.
Run the decay-driven ADLC maintenance checks โ stale skills, hot files to re-prosecute, stale tickets, and gate calibration (C10/C12).
Run the multi-lens P5 adversarial prosecution fan-out and record ADLC P5 review evidence and P6 acceptance packet workflows in the GitHub Copilot CLI. Fans out five independent read-only lenses, dedupes findings, verifies each independently, and loops until dry before recording evidence.
| name | adlc |
| description | Route software work through the Agentic Development Lifecycle in Cursor. Use when the user asks to apply ADLC, operate in ADLC, triage work, choose the next ADLC gate, or mentions P0โP7 / rails / prosecute. |
Treat agentic development as phases with machine-checkable gates via
adlc <tool> (exit 0 = pass, 1 = operational error, 2 = gate fail).
Prerequisite: adlc --version works (npm i -g @adlc/cli). Run /adlc-init
or adlc init --harness cursor once per repo for the .adlc/ workspace.
Vague request, no ticket yet? โโโโโโโโโโโโโโโโ P0 /adlc-ticket
Have a spec / acceptance criteria? โโโโโโโโโโโ P1 /adlc-spec ยท /adlc-approve-spec
Have tickets, planning fan-out? โโโโโโโโโโโโโโ P2 /adlc-decompose
About to build, want to freeze tests? โโโโโโโโ P3 rails-guard hooks + CI
Mid-build / verify? โโโโโโโโโโโโโโโโโโโโโโโโโโ P4 /adlc-verify-build
Change done, pre-merge prosecution? โโโโโโโโโโ P5 /adlc-prosecute
Repeated review findings to bank? โโโโโโโโโโโโ P7 /adlc-distill
Idle-time maintenance? โโโโโโโโโโโโโโโโโโโโโโโ โ /adlc-maintain
LLM-backed gates support --prompt-only: print the prompt, answer it yourself
in-session (no API keys). Prefer that over wiring providers.
Enforcement honesty: Cursor preToolUse deny is best-effort. The
unbypassable control is the CI rail-freeze gate (docs/ci/rails-guard.yml).
MCP tools (packaged): the plugin ships mcp.json โ a lifecycle Roots
proxy (bin/adlc-mcp-wrapper.mjs) that spawns adlc mcp-server only after
resolving a consumer root. Tools: adlc_gate, adlc_prosecute. Requires
adlc on PATH (npm i -g @adlc/cli). Status until an installed-Cursor Roots
proof is recorded: wrapper landed / channel unverified โ do not treat MCP
as production-shipped. Never guess process.cwd(); multi-root ambiguity fails
closed.
P5 caveat: prefer Task fan-out via packaged agents/prosecutor-* (fresh
context per lens). Sequential same-context is a degraded fallback with
weaker independence. For the cross-model risk gate run
npx adversarial-review --providers <a,b>.