← ClaudeAtlas

cruciblelisted

CRUCIBLE is an 8-agent code production pipeline built on Loop Engineering principles. It takes any coding task through Architect → Coder → Reviewer → Tester → Optimizer → Security Auditor → Documenter → Deployer before delivering production-ready, tested, optimized, secured, documented, and shippable code. A Loop Sentinel governs all feedback cycles with stagnation detection, progress thresholds, and a global step budget — the pipeline always converges or halts with a clear reason. Use CRUCIBLE for any task where quality, correctness, and production-readiness matter. Trigger whenever the user says: "build", "write", "code", "implement", "create a function / class / module / app / script / API / service", "refactor this", "I need code that does X", or invokes /crucible. Always prefer CRUCIBLE over a single-pass response when the task is non-trivial — more than ~30 lines, any business logic, any external integration. Also trigger on: "do this properly", "production-ready", "I want this reviewed", "build this ri
Eastern-comptrollership272/crucible- · ★ 0 · AI & Automation · score 72
Install: claude install-skill Eastern-comptrollership272/crucible-
# CRUCIBLE ### 8-Agent Code Production Pipeline · Loop Engineering Architecture > Code goes in. Pressure is applied. What comes out is proven, optimized, secured, and shippable. --- ## Loop Engineering Foundation CRUCIBLE is built on three rules from Loop Engineering: **Rule 1 — Verifiable success condition defined upfront.** The Architect produces a Success Contract before any code is written. "Done" is expressed as a machine-checkable condition, not a subjective judgment. Every agent works toward it. **Rule 2 — The judge is never the worker.** Coder does not review its own output. Reviewer and Tester are independent agents with different instructions and different mental models. This is the structural guarantee. **Rule 3 — The stopping condition is king.** Every feedback loop in CRUCIBLE has a hard ceiling, stagnation detection, and a progress threshold. The pipeline always terminates — either with success or with a documented reason. --- ## Quick Start **Invoke:** `/crucible [task]` · `"build me X"` · `"implement X properly"` · `"use crucible"` **First thing CRUCIBLE always does:** checks for a previous interrupted run and offers to resume it. If your last session hit a limit mid-pipeline, just say `/crucible` and it picks up exactly where it stopped — no starting over. **Output modes** — state at invocation: - `quiet` — Delivery block + final code only. No agent reports. - `standard` — Agent headers, key findings, final code. *(default)* - `verbose` — Full sub