agents-md

Reference: https://harn.app/kb/specs.html — "12 Factor Agents"

Reference: https://harn.app/kb/context.html — "Writing a good CLAUDE.md"

AGENTS.md Builder

Generate a lean, effective AGENTS.md using progressive disclosure principles.

Rules

1. Under 60 lines — ideally under 100, never over 300
2. No directory trees — agents discover structure on their own
3. No bloated context — point to skill files for domain-specific rules
4. Progressive disclosure — load detailed instructions only when relevant

Process

1. Read the current project root (package.json, README, existing AGENTS.md)
2. Identify: stack, entry points, architectural constraints
3. Generate AGENTS.md with sections:
   • North Star (1-2 sentences)
   • System of Record (stack, entry points)
   • Constraints (hard rules)
   • Active Harness Hooks (what's automated)
   • Skills (progressive disclosure pointers)
   • Escape Hatch (what to do when stuck)
4. If existing AGENTS.md is over 100 lines, refactor it:
   • Extract domain rules into separate skill files
   • Keep only pointers in AGENTS.md

Workflow

1. Understand → Read code, check LEARNED.md for gotchas
2. Plan → Break task into steps, update CHECKPOINT.json
3. Implement → Write code within architectural constraints
4. Verify → Run quality gate before finishing
5. Document → Update LEARNED.md if something was tricky

Context Budget

• Keep this file under 60 lines — load skills/ on demand
• Delegate research to sub-agents — only summaries return
• After 30+ tool calls, compact and update CHECKPOINT.json

Escape Hatches

• Quality gate stuck? Stop hook checks stop_hook_active — retry lets you through
• Security guard wrong? Report false positive, use alternative command
• Same error 3 times? Stop and ask the human