anvil

Code-review an anvil task's submitted changes against its acceptance criteria; return a structured PASS / SHOULD FIX / MUST FIX verdict. Read-only. Triggers: "review the changes for <task>", "review against acceptance criteria", "is this ready to accept".

Maintain anvil's inward-facing documentation: the `docs/` folder (specs, runbooks, design notes, plan archives), `CHANGELOG.md`, and the `description` field of `.claude-plugin/plugin.json`. Audits cross-references — broken `[[wikilinks]]`, mismatched anchors, dangling `see also` pointers, moved/archived file references — and reconciles prose against the actual code (schema, CLI, agents). Triggers: "update anvil docs", "fix broken links", "write the changelog", "doc cross-reference audit", "after-phase docs sweep".

Turn a parsed anvil PRD into a coherent task graph — group Requirements into Features, draft Tasks with acceptance criteria and verification commands, flag high-complexity tasks for expansion. Proposes; never writes state. Triggers: "generate features and tasks", "plan from the PRD", "extend the task graph for new requirements", "break down / expand this task".

Validate that submitted evidence on an anvil task actually proves the acceptance criteria — re-run verification commands, inspect outputs, return a binary PASS / FAIL scorecard. Read-only. Triggers: "verify the evidence for <task>", "does the evidence prove the criteria", "re-run verification". Unlike critic (code quality), sentinel checks evidence completeness.

Run sync reconciliation across anvil's three sources of truth inside an initialized project — SQLite (.anvil/state.db), the filesystem (packets/, .evidence-buffer/, worktrees), and git (branches, commits, claims). Surfaces drift as a structured report (orphan branches, orphan packets, stale claims, missing sync_mappings); reports only, never remediates — that is the user's choice via `anvil sync --fix --yes`. Triggers: "reconcile state", "sync drift", "check for orphans", "audit anvil", "is my project state stale".

Acquire an exclusive lease on an anvil task — pick from the ready queue, check for file conflicts, claim the task, and get a working git branch to commit into. Use this skill when ready to start work on an approved task.

Run the agentic execution loop on a claimed anvil task — fetch the work packet, do the work, submit completion evidence. Use this skill when an agent has just claimed a task and needs to execute it end-to-end without juggling individual CLI commands.

Decide what to do with an anvil task that has submitted evidence and is awaiting human review — accept and ship, reject and reopen, or hold for further investigation. Use this skill when one or more tasks are in needs_review and need a final disposition.

Turn a reviewed PRD into a ready-to-execute task graph — generate features and tasks, score each on six dimensions, surface dependencies and conflict groups, promote drafted tasks to ready. Use this skill once the PRD is approved and before any agent claims work.

Author, parse, and review a project PRD in anvil — capture the requirements that everything downstream (features, tasks, claims, evidence) gets generated from. Use this skill when starting a new project or revising requirements before any planning work happens.

4 hooks across 3 events