adna AGENTS.md

adna — Agent Guide

Purpose

This is the root agent reference for the adna project. It routes agents to the appropriate triad leg (what/how/who) based on their current task. This file is always loaded — it does not have a Load/Skip section because agents need it for orientation on every session.

Quick Orientation

Document	What You'll Find
CLAUDE.md	Agent operating context — safety rules, startup checklist, protocols
MANIFEST.md	Project overview — architecture, entry points, active builds
STATE.md	Current operational state — phase, blockers, next steps
README.md	Human-readable getting-started guide

Project Structure

adna/
├── what/    # WHAT — Knowledge objects, context library, decisions, docs, lattices
├── how/     # HOW — Campaigns, missions, sessions, templates, pipelines, skills, backlog
└── who/     # WHO — Coordination, governance (extend with team/, customers/, etc.)

Navigation: Read this file → follow the Layer References below into the triad leg matching your task → drill into specific subdirectories as needed. Each AGENTS.md narrows context progressively.

Agent Startup

1. CLAUDE.md — auto-loaded; confirms project structure and rules
2. First-run check — if uncustomized vault, invoke onboarding skill (how/skills/skill_onboarding.md)
3. STATE.md — operational snapshot: current phase, blockers, next steps
4. how/sessions/active/ — check for conflicting sessions
5. who/coordination/ — read any urgent cross-agent notes
6. how/backlog/ — quick scan for ideas relevant to current session
7. how/campaigns/ — check for active campaigns
8. how/missions/ — check for active missions
9. Create session file in how/sessions/active/ and begin work

Layer References

• what/AGENTS — Knowledge objects (WHAT)
• how/AGENTS — Operations (HOW)
• who/AGENTS — Organization (WHO)