Claude Code is RAC's most integrated client — it gets the two surfaces every client gets (a generated context file + the lore MCP server), plus two Claude-Code-specific extras: a bundled authoring skill and the only platform seam that allows a real pre-edit veto. A stranger can reproduce this from the file alone.

pip install rac-core   # the `rac` CLI and the `lore` MCP server

A repository with a RAC corpus under decisions/ (run decided quickstart, or use this repository's own decisions/).

decided export decisions/ --agent-rules

Writes the settled decisions into a managed block in CLAUDE.md at the repo root (your own content outside the block is preserved). Claude Code reads CLAUDE.md automatically. Re-run on change; decided export decisions/ --agent-rules --check fails CI if it drifts.

claude mcp add lore -- decided-mcp --root .

…or commit a project-level .mcp.json so the team shares it:

{
  "mcpServers": {
    "asdecided": { "command": "decided-mcp", "args": ["--root", "."] }
  }
}

Exposes the five read-only tools get_summary, search_artifacts, get_artifact, get_related, find_decisions. The server re-reads the corpus on every call and never writes to the repo.

decided skill install rac-artifacts

Installs a project skill that teaches Claude Code to create, validate, and update RAC artifacts with the rac CLI (it only touches the decisions/ subtree). decided skill list shows the bundled skills (rac-artifacts, rac-import, rac-ingest, rac-review).

RAC supplies context and enforces after the edit (ADR-067); it does not rewrite Claude Code's loop. Two optional guards:

• Git hook (any client). decided hook install --style pre-commit validates staged artifacts on commit (--style post-commit is an advisory cadence nudge that never blocks). This is a git hook, not a Claude Code hook.
• Pre-edit veto (Claude-Code-only). Claude Code's PreToolUse hook is the one platform seam that can block an edit before it lands. The RAC VS Code / Cursor extension generates it ("RAC: Enable Claude Code pre-edit hook"), or you can register it by hand in .claude/settings.json under hooks.PreToolUse: it pipes the proposed content to decided validate - --corpus decisions/ and blocks (exit 2) only on a structural finding — a reference to a retired or missing decision, or a malformed artifact — and fails open on any internal error. All validation stays in rac; the hook computes nothing (ADR-063, ADR-067).

Either way, the CI / PR gate (decided validate, decided relationships --validate) remains the backstop, regardless of which agent edited.

Run the bundled grounding demo — same task twice, once unconnected and once with lore connected — and watch the connected run respect a recorded decision the unconnected run violates: examples/guide/.