Ringmaster

Resume Ringmaster work from the .ringmaster/ ledger — reconcile against the repo, then continue the highest-priority pending task.

A fresh-context comprehension examiner that turns an AI-written change into active-recall questions grounded in the actual diff, so the developer reconstructs real understanding and can honestly take 100% ownership before sign-off. Dispatch it in PHASE 1 to read the working/staged diff, risk-tier it, and produce a grounded question bank across five levels (architectural · code · functional · business · test) — each question carrying an evidence anchor, a model answer, the load-bearing points that must appear, and the common wrong turns. The orchestrator then conducts the quiz in the main thread (answer-first), grades against the anchors, reveals and teaches on any miss, and emits a Comprehension Record for the sign-off gate. It never fixes code, never commits, and NEVER asserts an answer it cannot ground in the diff or evidence.

A fresh-context adversarial security reviewer. Dispatch this subagent on the working or staged diff BEFORE staging/hand-off on any change that touches production behavior — features, frontend with user input, database/query changes, auth, payments, anything handling secrets or external data. It hunts for vulnerabilities, secrets, injection, broken authorization, crypto misuse, and risky dependencies; it BLOCKS the hand-off on any critical finding. It reports defects in a compact, plain-language report and hands them back — it never edits code to fix them (the human decides), and never commits or pushes.

The decision-making ringmaster for any software task. Use at the START of essentially any build, change, fix, or review request, and whenever a request is vague ("help me build X", "add Y", "fix this", "make it better"). Covers new/updated frontends, new/changed features, writing/updating/pruning tests, databases, payments, current-library lookups, refactors, docs, and skill/plugin creation. It frames the real goal, plans in plain-language chunks behind an approval gate, routes each step to the best available specialist or a built-in fallback, and runs the bundled Test Architect and adversarial Security Gate before handing off staged, reviewed work. Enforces hard safety rails: never commits/pushes/merges, never runs against production. Stack-agnostic. Trigger even when no specific tool or test framework is named.

The ownership review - use when a developer must take 100% responsibility for a code change written largely by AI and wants to genuinely understand and stand behind it before shipping. Trigger on 'review this branch/PR', 'code review', 'help me own this code', 'I have to be responsible for this', 'sign off on this change', or 'make sure I understand this before I ship'. It runs AFTER detection (the Security Gate and the code-review pipeline find the bugs) and adds what they cannot - it reconstructs the developer's understanding through active recall by dispatching the comprehension agent to generate diff-grounded questions, conducts the quiz answer-first in the main thread, teaches on every miss with grounded explanations a junior could follow, and records an auditable ownership sign-off (gate.owned) that the Stop hook enforces. It never fixes code, never commits, and never rubber-stamps.

Acts as a Principal Test Architect to plan, write, prune, run, and hand off automated tests for any codebase in any language or framework — never touching production code, never committing, never running against production. Use whenever the user wants to add, fix, expand, refactor, or review tests for a feature, bug, ticket, or component — including TDD / red-green work, regression coverage, raising test quality, hardening flaky suites, or pruning stale tests, behind an explicit plan-and-approve step with a clean staged handoff. Trigger even for "write tests for X", "cover this with tests", "add a test for this bug", "tighten up our test suite", or a named ticket — and even when no test framework is mentioned.

5 hooks across 3 events