| name | self_improve |
| description | Real-time reflex for improving yourself — turn user feedback, preferences, and workflow patterns into memories and skills as they happen |
| when_to_use | After any feedback moment in a conversation — a correction, a stated preference, a missed pattern, a better flow discovered. Do NOT defer to end of session. |
| created | 2026-08-05T09:16:13Z |
| source | chat |
Self-Improve — Real-Time Reflex
Run this checklist immediately after any feedback moment. Do NOT accumulate and review at end of session (token-expensive and easy to forget).
Checklist
1. Is this a user preference or stable fact?
- e.g. "always show controllers before asking for params", "use 3 months as default backtest window"
- →
manage_memory(action="write", ...) — one memory per fact, type=preference
- Memory is YOURS ALONE — keyed by (assistant, user), so what you write here
is invisible to the chat and to every other agent, and there is no way to
publish it. Write it anyway for your own work, but if the preference should
govern how everything behaves, say so to the user — a cross-cutting rule
belongs in a shared skill (below), not in one agent's memory.
2. Is this a workflow pattern or missed step?
- e.g. "when asked to backtest, list available controllers first"
- → First:
manage_skill(action="list") — check if an existing skill covers it
- If yes:
manage_skill(action="read", name=...) to confirm it is the same
pattern, then manage_skill(action="edit", ...) — update the existing one
- If the read returns
inherited: true, the playbook lives in the
shared library and is read-only for you. edit will fail. To specialize
it, create a skill with the SAME name — it shadows the shared one for
you only. If the improvement is good for everyone, ask the user to have
the chat edit the published version instead of shadowing it.
- If no:
manage_skill(action="create", ...) — create a new one
- Whose library? A skill is written to the CALLER's library. Before creating,
decide who the pattern belongs to:
- It is about how you work → create it normally (no
agent argument).
- You are the chat and the pattern belongs to a domain agent → pass
agent="<agent_slug>" so it lands in that agent's library. Skipping this is
the #1 cause of skills ending up on the wrong agent.
- It should apply to ALL agents → only the chat can publish: create it with
shared=True, which puts it in the shared library every assistant reads.
3. Is this a one-off or session-specific detail?
- → Skip it. Don't pollute memory/skills with ephemeral context.
Rules
- Save in the moment, not at the end
- One memory = one fact (no bundling)
- Always check existing skills before creating a new one
- Decide the OWNER before writing: memory is per-agent and cannot be shared;
skills are per-agent but the chat can publish one to all with
shared=True
- Shared skills land in every agent's context — publish deliberately
