agent-assetslisted

# Agent assets audit Everything an agent needs that isn't application code: the reusable commands, the scheduled workflows, the vendored reference material, the tool configuration, the orientation files. These accumulate without anyone owning them, drift out of sync with the code, and fail quietly. The question throughout: **is the knowledge that makes agents effective here committed and current, or is it in someone's terminal history?** A prompt that works but gets retyped each time is not an asset. Neither is a workflow that runs nightly and produces nothing anyone reads. Related skills: `agent-navigability` covers whether the code is legible, `agent-testability` whether work can be verified, `logging-audit` whether runtime output is usable. This one covers the scaffolding around all three. --- ## Step 1: Inventory what's there Look for these, and note their absence as readily as their contents: - `.claude/commands/`, `.cursor/rules/`, or equivalent reusable prompts - `CLAUDE.md`, `AGENTS.md`, and any subsystem-level equivalents - Vendored documentation — `ai_docs/`, `docs/vendor/` - Agent workflows in CI, and the scripts they call - MCP or tool configuration, and whether it's committed or per-developer - Committed plans or specs, and whether the code reflects them Then the check that finds the most: **open each one and ask whether it still describes this repo.** Stale assets are worse than missing ones — they produce confident wrong work, and they survive because n