akashic-memorylisted

# Akashic memory: the loop you are inside This repo has a shared memory. Other agents' lessons surface to you automatically (hooks inject the top few before risky actions); what YOU learn must flow back or the loop starves. The full contract is `AGENTS.md`; this skill is the reflex layer. ## Session start ``` py agent_cli.py boot <your_agent_id> --task "<this slice only>" ``` Read the output. The RECENT NOTES section is where-we-are; `py agent_cli.py notes --json` for full bodies. Never re-read chat history to reconstruct state — the store is the continuity layer, the chat is disposable. ## The capture reflexes (highest value, most forgettable) **FAIL→SUCCESS flip** — the moment something that failed now works, a lesson was just earned. The hook usually nudges you with a pre-filled command; run it. Write the recommendation TRIGGER-PHRASED: ``` py agent_cli.py learn <id> --experiment <slug> \ --tried "<what failed>" --result "<what fixed it>" \ --recommend "Use when <symptom>, before <action>: <advice>. Don't when <contraindication>." ``` Include what did NOT work (`--tried` is exactly that) — failed approaches save the next agent more time than successes do. **User correction** — every time the human corrects you, that is a lesson-earning moment (the creator of Claude Code runs this reflex manually on CLAUDE.md; here it has a door). Record it immediately with `--category correction`. Do not just comply and move on. **Known-bad approach** — record with `--anti-