loop-library

# Loop Library A loop = an autonomous iterative workflow with a hard **stopping condition** and **evidence-backed output**. The shape is always: > **GOAL** (one concrete outcome) → **PROCESS** (measurable steps, each verified) → **STOP** (explicit halt criteria, including a max-iterations / dry-rounds backstop) → **OUTPUT** (documented result + proof). ## How to run a loop here - **`/loop` (self-paced)**: let the model iterate until the stop condition; good for "fix until N pass" / "find until dry." Omit the interval for self-pacing. - **`/loop <interval>`**: recurring cadence (e.g. a maintainer loop every N hours). - **loop-until-done runner**: the canonical "keep going autonomously until done" runner. - **fan-out orchestrator**: when the loop fans out (loop-until-dry, loop-until-count across many items). - **scheduler (cron)**: for the recurring/overnight ones. Every loop MUST declare its stop + a backstop (max iterations or K consecutive empty rounds) so it can't run forever. Every loop ends by logging its output + evidence. --- ## 1. Docs sweep (engineering) - **Goal:** your repo docs (CLAUDE.md, skills/_index.md, knowledge/, memory/MEMORY.md pointers, READMEs) reflect the current implementation. - **Process:** scan for drift (a skill/script/cron that exists but isn't documented, or documented-but-removed). For each drift: fix the doc, verify the referenced file/flag still exists (don't trust the memo, check the file). - **Stop:** zero drift found on a full pass, OR...