Anchorwatch Pro

Guardrails stop the damage. Pro adds the discipline.

The free plugin stops Claude from doing damage. Pro makes Claude work the way a careful senior engineer works: format everything, run the tests, write real commit messages, get a review before merging, and never lose the thread across a long session.

What each plugin actually does

Quality Gates

A PostToolUse hook runs the formatter your repo already uses (prettier, biome, ruff, black, gofmt, rustfmt — auto-detected, nothing runs if you have none) on every file Claude edits, then scans non-test code for console.log, debugger, print(, dbg!, binding.pry. A Stop hook checks whether code changed since the last test run and whether that run passed; in remind mode it tells you and Claude, in enforce mode it blocks the first attempt to stop so Claude goes and runs the tests. /quality-gates:preflight runs lint → typecheck → tests → build and gives you a scorecard.

Ship

/ship:commit reads the actual diff, splits unrelated changes, and writes Conventional Commits — never --no-verify, never git add -A. /ship:pr pushes, fills your PR template (or a good default: summary, changes, test plan, risks), and links issues. /ship:release runs gates (clean tree, default branch, tests green), infers the semver bump from the commits, writes the CHANGELOG section, bumps every version file, and only after you confirm: commits, tags, pushes, and creates the GitHub release.

Review Crew

Four subagents — security-reviewer, perf-reviewer, test-gap-finder, contract-reviewer — each with a detailed checklist and a strict output format (severity, file:line, evidence, failure scenario, fix). /review-crew:review runs them in parallel on your working tree, a git range, or a PR number, de-duplicates and ranks the findings, and gives a verdict. --fix applies the critical and high fixes and re-runs the tests.

Context Keeper

A PreCompact hook writes a snapshot (branch, uncommitted changes, diff stat, recent commits, recently modified files, your HANDOFF.md) and a SessionStart hook re-injects it after compaction or resume, so Claude doesn't "forget" what it was doing halfway through. /context-keeper:handoff writes a precise HANDOFF.md; /context-keeper:resume reconciles it with git and proposes the next step.

Setup Audit

/setup-audit:run inventories CLAUDE.md, .claude/settings.json (secrets masked), hooks, rules, skills, agents, MCP servers, memory, and repo hygiene (including a spot-check of git history for leaked keys), grades each area A–F against a published rubric, and lists the top five fixes ordered by risk × ease. --apply makes the safe local ones.

Stack Packs

/stack-packs:init detects your stack and writes a CLAUDE.md filled with your real commands, package manager, versions and layout, plus path-scoped .claude/rules. If you already have a CLAUDE.md it merges and flags contradictions instead of overwriting.

FAQ

How is Pro delivered? After purchase Polar grants your GitHub account read access to the private anchorwatch-pro repository. You add it as a marketplace once (claude plugin marketplace add anchorwatch-dev/anchorwatch-pro) and install plugins normally; updates flow through the marketplace like any other plugin. Prefer a zip? It's in your Polar purchase too.

Which Claude Code version? Current releases (2.1+). On older 2.0 versions the free plugin still blocks the same commands, but the block surfaces as a permission prompt rather than a written reason Claude can act on. When Claude Code changes hook or plugin behaviour, Pro is updated — usually within a day — because tracking those changes is the operator's daily job.

Does it send my code anywhere? No. Everything runs locally in hooks and skills. This site collects anonymous pageviews (see privacy); the plugins collect nothing.

Refunds? 14 days, no questions, via Polar.

Team licences and invoices? The Team tier covers up to 10 developers; Polar issues VAT invoices automatically. Need more seats? Open an issue.