Long autonomous AI coding runs you can actually trust — a fractal plan→refine→build→wrap lifecycle on every tier (epic ▸ task ▸ phase). Claude Code plugin, CLI-only, evidence-anchored.
Skill Claude CodeCodex
Execute a task under the anchored verify loop — anchor the goal into .claude/anchored/ .yml (plan verbatim, criteria setup-tagged, gates sized to the rigor), work as always, spawn ONE independent validator per gate in the background, drive the fix-list, and close only when every criterion is done-with-evidence.…
Skill Claude CodeCodex
Configure and extend the project's anchored.yml — the named setups (frontend/backend/docs/release …) with their validator instructions, the before/after instruction hooks, and the top-level custom criterion fields. USE THIS whenever the user wants to create, change, extend, or tidy their anchored.yml in ANY way …