This file is the vendor-neutral operating manual for agents working in this repository.
CLAUDE.mddocs/REQUIREMENTS.mddocs/PLAN.mddocs/PROGRESS.mddocs/RUN_CONTEXT.mddocs/NEXT_STEPS.md
- Read project memory before planning or code changes.
- Clarify the active objective in
docs/REQUIREMENTS.md. - Update
docs/PLAN.mdbefore substantial implementation. - Track completed work, decisions, and validation in
docs/PROGRESS.md. - Keep
docs/NEXT_STEPS.mdcurrent as the handoff for the next session. - Use git worktrees for non-trivial parallel work.
- Use Conventional Commits for final commit messages.
agent-osis the source of truth for the project scaffold.- Local Python-backed
agent-oswork runs in Condabaseat/Users/junlee/miniconda3. - Homebrew Python is not the supported runtime for
agent-os.
- Prefer the smallest useful verification step first.
- Do not duplicate large prompt blocks in chat when they belong in project memory.
- Record environment limits, APIs, and rate limits in
docs/RUN_CONTEXT.md. - Keep
CLAUDE.mdshort and use it as an index into the live docs.