← ClaudeAtlas

myna-dev-improvelisted

Full Myna quality pipeline — lint until clean, then run review → fix → verify cycles until issues are resolved or cycle limit is reached. Long-running. Use when: "improve the skills", "run quality pipeline", "full QA pass", "lint and review".
agentflock/myna · ★ 4 · AI & Automation · score 83
Install: claude install-skill agentflock/myna
# Myna Dev Improve Run the full Myna quality pipeline end-to-end: lint until clean, then run Review/Fix/Verify cycles until issues are resolved or the cycle limit is reached. Long-running — every phase gets full attention. **Role:** Engineering Manager — orchestrates the full pipeline, monitors progress, knows when to stop early, and prevents thrashing. ## Arguments $ARGUMENTS Parse for: - `--cycles N` — max Review/Fix/Verify cycles (default: 3, min: 1, max: 10) - **Scope:** file paths, glob patterns, or `--uncommitted` - No scope → all feature skills + main.md - `--include-steering` — also include steering skills - Anything else: error out ## Pipeline ### Phase 0 — Lint Run `bash scripts/lint-agents.sh`. Fix all FAIL items. Re-run until clean. Lint errors are blocking — no review starts until lint passes. ### Phase 1-N — Review/Fix/Verify Cycles Each cycle: 1. **Review** — spawn one subagent per skill (parallel for 4+ files). Evaluate against 10 dimensions (frontmatter, description, instruction clarity, feature coverage, vault formats, safety, edge cases, output, steering duplication, conciseness). Write report to `tmp/reviews/review-{NNN}.md`. 2. **Fix** — read review report, implement all fixes (Critical → Important → Minor → Nitpick). Cross-skill consistency check after fixes. Write report to `tmp/reviews/fix-{NNN}.md`. 3. **Verify** — confirm each fix resolved the original issue. Regression check on all modified files. Write report to `tmp/reviews/verify-{NNN}.m