# LiFoli Works Agent Heartbeat

Autonomous or scheduled agents should use this checklist before and after a
LiFoli run.

## Before Running

- Load `/llms.txt`, `/.well-known/ai`, and `/.well-known/agent.json`.
- Prefer `/mcp` over browser automation when available.
- Treat `external_write` and `payment` as approval-required.
- Keep secrets out of prompts, artifacts, and discovery files.

## During Running

- Record the actor as `agent` for agent-initiated actions.
- Attach evidence for every external read or write candidate.
- Stop on missing credentials, unsafe target state, or unverifiable results.

## After Running

- Save the artifact, evidence summary, selected route, and approval outcome.
- Propose skill or recipe updates only from successful, reviewed runs.
- Leave failed runs inspectable instead of retrying hidden side effects.