Midas — a portable product-development harness for AI agents: drive a product from idea to shipped code through 9 audited phases, with cost-tiered agents, Context7, and a whole-project adversarial debate (/midas-tribunal).
This repository also configures its own agents. See what midas-harness tells them →
Skill Claude Code
Run diff-scoped test/quality gate receipts into {paths.cache}/gates/ /; use before close-sprint when production paths changed. Does not replace close-sprint or midas-verify.
Skill Claude Code
Over-engineering review — delete-list for the current diff or named paths (stdlib/native/yagni/shrink). Use before /close-sprint on a fat diff, after a feature branch, or when the user asks what to cut. Complements correctness review; does not replace /midas-sweep or /close-sprint.
Skill Claude Code
Phase 7 STM writer — updates {runs}/sprints/NN-progress.md after tasks or significant decisions (Done rows, Learned observations, Next line). Build-tier; complements read-only /midas-recall. Use mid-sprint when session-continuity rule applies.
Skill Claude Code
Ad-hoc branch/PR QA — map diff to routes/screens, drive with agent-browser (web) or Maestro MCP (native), report in chat; optional {runs}/qa/qa-adhoc-.md. Use during Phase 7 inner loop; does not replace /midas-verify before /close-sprint.
Skill Claude Code
Hygiene and dead-flow sweep — find orphans, stale docs, ledger drift, and harness mismatches; freeze to {runs}/sweeps/sweep-NN.md. Optional --fix with explicit confirm. Use on demand, after brownfield adopt, or before closing a large sprint.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: