Memory

AGENTS.md is the only file humans review regularly — agents write and update it. docs/agents/ holds session/working context; do not duplicate AGENTS.md there.

On activate (start of session or when user invokes memory)

1. Read AGENTS.md — follow Purpose, Rules, and Tech stack before planning or editing
2. Read relevant docs/agent/*.md
3. Run AGENTS.md audit (below) — refresh sections when config, README, or chat changed
4. Self-check before finishing: re-read Purpose and Tech stack; run repo lint on touched files

AGENTS.md audit

1. Read repo config to see what's in use — check when relevant: package.json, .oxlintrc.json, vite.config.*, tsconfig.json, bunfig.toml, pyproject.toml, turbo.json, README, existing Purpose
2. Read the techstack skill for product-domain stack decisions and the tooling skill for lint/format/build/verification conventions
3. Build Tech stack from techstack + tooling defaults that apply here + repo-specific facts config alone doesn't convey

Build Tech stack from techstack and tooling