mdownreview

Use when the user asks about benchmark trends, says "perf trend", "/perf-trend", "show bench history", "check if perf is regressing", or "compare benches across commits". Reads the `bench-output` artifact from recent successful `ci.yml` runs on `main`, parses criterion bencher output, and renders a markdown report with per-bench history, budget status, and regression callouts.

Use when the user says "release", "tag", "publish", "ship a new version", "bump version", or runs `/publish-release [major|minor|patch]`. Fully autonomous — never prompts. Default bump: patch.

Use when the user wants the agent to ship the iterate PR backlog autonomously — phrases like "merge the ready PRs", "ship the iterate PRs", "watch CI on the ready PRs", or an empty `/merge-pr-loop`. Continuous (default) or single-pass (`--once`). Picks the next open ready-for-review PR labelled `iterate-pr`, **rebases stale PR branches onto `origin/main` (clean rebases inline; conflicts delegate to `/iterate-one-issue --resume-pr`)**, waits for the auto-triggered CI run on the PR's HEAD, drives forward-fixes via `/iterate-one-issue --resume-pr <PR>` on failure (cap 5/PR), squash-merges on green. Never prompts. Never writes app source. Pair with `/iterate-loop` in another terminal — that loop produces the PRs this one ships.

Use when verifying that CI runs on the current work or when the user asks to "trigger CI" / "validate before merging" / "force a re-run on the same SHA". Since [PR

Use when the user wants the agent to drain the GitHub issue backlog autonomously — phrases like "drain the backlog", "work through the issues", "auto-fix open issues", or an empty `/iterate-loop`. Continuous (default) or single-pass (`--once`). Picks the next eligible open issue, dispatches `iterate-one-issue` synchronously in the foreground, releases the claim, repeats. Leaves each PR ready-for-review with the `iterate-pr` label for `merge-pr-loop` to ship. Never prompts. Never merges. Pair with `/merge-pr-loop` in another terminal.

Use when the user asks to fix a specific GitHub issue, references one (`#42`, `issue-42`, an issue URL), or asks to pursue a freeform engineering goal — phrases like "work on issue 42", "implement

Use when the user runs `/groom-issues`, asks to "groom", "spec out", or "process the needs-grooming backlog" of GitHub issues, or pastes issue numbers asking for grooming. Two-phase pipeline (interactive Phase A → autonomous Phase B).

Use when the user asks for exploratory or "dogfood" end-to-end testing of the mdownreview app — phrases like "explore the app", "find UX bugs", "test the live build", or when invoked by the test-exploratory-loop skill. Fully autonomous, files deduplicated GitHub issues. Windows-only.

Use when the user asks to "dogfood the app continuously", run exploratory testing in a loop, or stress-test mdownreview while the iterate loop fixes the backlog. Fully autonomous (never prompts). Windows-only. Args documented in body.

Use before declaring work complete, before opening a PR, after rebasing, or whenever the user says "run the tests", "run the build", "verify it works locally", or "make sure CI will pass". Runs every build, lint, and test gate that can run on this machine.

Use when authoring or editing a skill (SKILL.md), subagent definition, slash command, system prompt, CLAUDE.md/AGENTS.md instruction block, or any reusable prompt the user is about to save. Trigger on phrases like "create/update skill", "write/edit agent", "tweak this prompt", "tighten this", or whenever the user pastes prompt-shaped text and asks for review. Rewrites for clarity, triggering accuracy, and token efficiency without losing intent.