TRACE

A playbook for AI-first software development. Adopt it in your repository to get a documentation structure, an opinionated plugin, and a set of conventions that keep an AI agent productive across a project's lifetime — without the process tax that usually comes with "AI governance."

Adopting TRACE in a fresh repo takes a few minutes and produces:

• A docs/ tree organized by what each file is: living description of the codebase (system/), prescriptive rules the system must follow (architecture/), and historical decisions and rationale (adr/, reference/, working-notes/).
• A root AGENTS.md describing your project's conventions (one source of truth, loaded automatically by Claude Code, Cursor, Codex).
• The playbook Claude Code plugin installed, which automates the parts of the workflow most often skipped: writing the initial context file, scaffolding starter docs, planning substantial work with explicit acceptance criteria, recording decisions as ADRs, and distilling durable knowledge as work happens.

Where to go next

• Adopting TRACE in your project? Read deliverable/README.md. It's the three-step walkthrough — both readable by humans and executable by an AI agent. No clone needed; the doc structure ships inside the plugin, and the walkthrough can be followed directly from GitHub.
• Working on TRACE itself? Go to AGENTS.md. Different audience, different rules.

If you're an AI agent that was pointed at this URL: first check that you are not running inside TRACE's own source tree (the cwd does not contain deliverable/playbook/ at its root). If you are, stop and tell the user to cd to their target project. Otherwise, fetch deliverable/README.md and continue from there.