CLAUDE.md / AGENTS.md — Root Context Template

Summary

The single entry point every agent reads first. States the mission, business context, where state lives, and points to the other operating files. One root per agent (or one shared root + per-agent overrides). See AGENTS.md — root context.

Template

# CLAUDE.md — <agent or pipeline name>
version: 0.1
last_updated: 2026-06-05
 
## Mission
<one sentence: what this agent exists to do>
e.g. "Produce daily on-brand peptide-education slide carousels for the 10 niche TikTok accounts and funnel viewers to the website."
 
## Business context
- Entity: Market → Distribution (Growth)   # see Three Entities
- Channel: TikTok (niche knowledge accounts)
- Objective + proxy: optimize control-panel inputs (retention, hook rate, shares, saves) — NOT follower count
  (see Governing Artifacts → Scoreboard vs control panel)
 
## Where state lives
- Tracker: <path or DB>   (status lifecycle: idea → briefed → produced → scheduled → posted → measured)
- Content output: /content/<account>/<date>/
- Logs: /logs/<agent>/<date>
- Knowledge base (RAG): <path>
 
## Autonomy
- Level: L2 (approval-gated)        # L0–L4, see BAU vs Pilot
- Safety gate: human approves before publish (banned words / health claims)
- Performance tuning: weekly, not per-item
 
## The other operating files (load in this order)
1. guardrail.md     — definition of ready (pre-flight)
2. policies.md      — hard rules + conventions
3. skills.md        — how-to procedures
4. eval.yaml        — definition of done + golden examples
5. CHANGELOG.md     — what changed recently
 
## Hard rules (the non-negotiables, also in policies.md)
- Never publish autonomously. Never make a health/medical claim. Always include disclaimer.
- Never act without completing the guardrail pre-flight.

Keep it short and stable

CLAUDE.md is the map, not the detail. Detail lives in policies/skills/eval. If you find yourself editing CLAUDE.md often, the churny content probably belongs in one of the other files.

Related: policies.template · guardrail.template · eval.template · Part A1 - Agent Operating Files