AribSudia

arib-engine

AribSudia/claude-code-methodology/.claude/skills/arib-engine/SKILL.md

Engine | Autonomous engineering & product-led-growth campaign engine. Use when the operator hands over ownership and wants continuous, self-running improvement of a codebase — triggers include "take over as the expert", "make it all", "rework until you decide it's done", "run the campaign", "harden / audit / sweep the codebase", "keep shipping value autonomously", or any request to autonomously discover, fix, verify, and ship many improvements across CI-green PRs until the evidence says done. It runs adversarial multi-dimensional discovery sweeps (find→refute→confirm), ships one focused reversible PR per concern behind hard verification gates, applies explicit ship/escalate/decline frameworks, records durable memory, and concludes on an evidence-based closure test (handing the operator a decision list). STANDALONE by default (runs the whole methodology self-contained); orchestrates the broader arib-* family ONLY on explicit opt-in (`--with-arib-family`, or "leverage the arib family"). Self-pacing is delegated to Anthropic's `/loop`. Merge is AUTO by default — gated on all-green blocking checks AND a `verification-agent` RECONCILED verdict (intent ↔ actual change), with high-stakes classes (money/auth/compliance/secrets/breaking-migration) always held for a human; `--hold-merge` holds every PR for review. Prefer for sustained campaigns; for a single quick fix, just do the fix. Full reference: reference/AUTONOMOUS_ENGINEERING_METHODOLOGY.md.