Claude Mates

Claude Mates

Autonomous maintenance agents for your codebase, shipped as a GitHub composite action.

Each mate is a specialized Claude Code agent that runs on a schedule, analyzes your repo, and either files an issue or opens a PR with fixes. Mates never merge — humans decide.

Inspired by [AI Daemons](https://ai-daemons.com/spec/), built on [Claude Code](https://docs.anthropic.com/en/docs/claude-code).

The mates

Two classes:

**Drift mates** — nightly state-query over HEAD. Each is a Claude Code prompt tuned for one concern.

Mate	Purpose	Default model	Trigger
docs	Documentation quality, staleness, drift	Haiku	nightly
tests	Outdated/redundant tests, weak assertions	Haiku	nightly
dead-code	Unused symbols, orphaned files	Haiku	nightly
logic	TODOs, deprecated APIs, hardcoded values	Haiku	nightly

**Policy gate** — PR-scoped, wraps a specialized tool.

Mate	Purpose	Engine	Trigger
security	Security review, diff-aware, inline PR comments	`anthropics/claude-code-security-review` (Opus 4.1 default)	pull_request only

The security mate is a thin wrapper over Anthropic's specialized scanner: you keep the `mate: security` interface, the scanner does the analysis. See [examples/README.md](examples/README.md) for the PR-scoped invocation pattern.

**Key architectural difference from the drift mates**:

	Drift mates	Security mate
Findings go to	Job Summary panel (ephemeral, per-run)	Inline PR review comments (during PR lifecycle)
Opens a PR?	Yes, when there's a concrete fix	No — detection-only
Opens an issue?	No (avoids tracker noise at nightly scale)	Only when a PR merges with residual findings (via optional security-aftermath.yml companion). Rare × high-signal.
Blocks merge?	Never (advisory only)	Yes, when security is added as a re