owns when colocated knowledge must move with source changes. This skill owns how to write and structure that knowledge.

Load for where each layer lives and what it holds. Load if it isn't already loaded.

This skill is about how to write and maintain the directory-local pair: AGENTS.md and . The pair governs any tree agents work in — code, the KB, docs, work directories. AGENTS.md loads into an agent's bounded context as standing instructions: write it like a prompt, minimal, every line load-bearing.

Agents read AGENTS.md before opening anything else in the tree — write for that moment. Ask: what must someone understand before working here? That's what AGENTS.md captures.

Keep AGENTS.md as short as the directory allows, rarely past 200 lines. Include only what has substance:

An agent that only reads AGENTS.md should be able to work correctly here. An agent that also reads .context/ should be able to change things safely.

Reference depth, co-located with the code it describes. Where an agent goes when it needs contracts, architecture, or rationale in detail.

Sections (use only those with substance):

The directory is extensible: additional files alongside CONTEXT.md for specialized concerns.

Apply the every-session test: root AGENTS.md loads on every session. If knowledge is only relevant when working in a specific domain, it belongs in that domain's AGENTS.md or .context/, not root.

Apply the think-vs-lookup test: text whose removal would cause a wrong decision belongs in AGENTS.md. Text an agent would merely have to look up belongs in .context/. Text that changes no behavior gets deleted — agents already know how to code and follow common conventions.

Specific failure modes:

Claude harnesses read CLAUDE.md, not AGENTS.md. Give every AGENTS.md a sibling CLAUDE.md whose first line is — normally the whole file. Run on the containing tree after creating or moving AGENTS.md files: it creates missing mirrors, skips exact ones, and reports anything else as a conflict.

Never write shared instructions into CLAUDE.md. Claude-only knowledge is rare; when it exists, put it below the import and expect to keep flagging the file, so the divergence stays visible.

Loading differs by level. At the root, each harness auto-loads its own file every session: Claude reads CLAUDE.md, others read AGENTS.md. In subdirectories, Claude auto-injects CLAUDE.md when it touches files there; other agents see nested AGENTS.md only by reading it on entry. Don't lean on Claude's auto-injection: a nested AGENTS.md carries the local additions an agent needs on entry, with everything else inherited from ancestors.