maestro

Independently re-check a conclusion, fix, or design by spawning an agent prompted to refute it — not affirm it

Multi-agent deep audit of the current project — adversarially verified findings + prioritized roadmap, published to docs/deep-analysis/

Run a large, multi-place, uncertain task to completion — plan, adversarial review, serial execute, synthesize & verify

Implement a feature, fix, or whole app with the full disciplined process — clarify, investigate, plan, build, verify, deliver

Show one feature's whole journey across the suite — idea → PRD → design → tasks → verified → regression-clean → shipped — by composing the existing per-plugin views (read-only; no new store)

Use this agent to execute ONE well-scoped implementation task end-to-end (a feature slice, bugfix, refactor step, or test batch) after the scope is already clear. Ideal for parallelizing independent slices of a larger build. Do NOT use it for ambiguous asks that still need owner decisions — clarify in the main conversation first, then delegate with a precise brief.

The discipline behind maestro:regression — prove a change didn't break existing work by running the suite, an adversarial bounded blast-radius probe against documented architecture contracts, and honest coverage, then emit a verdict taskmanager's done-gate enforces. Trigger: a change is about to be marked done / committed and you must confirm it didn't regress existing behavior or violate a documented boundary.

Turn the work just done into VETTED lessons that survive across sessions and projects — harvest the session's decisions/errors, draft candidate lessons, refute them adversarially, and promote only the survivors into taskmanager memories. The self-improvement loop, gated so a false lesson can never silently change how the suite works. Trigger: end of a substantial piece of work / a wrap-up moment, or when asked to capture lessons learned.

The suite's risk-sized front door: given a request, name the right top-level PIPELINE (which existing entry points, in what order) by scope and risk — analysis, trivial change, bug fix, feature/high-risk, or codebase health. Advisory; it routes the pipeline only and never sizes verification depth (the brain's job). Trigger: a "which pipeline / where do I start / what process for this / route this" question, or an ambiguous ask whose right path is unclear.

Use when implementing any change in existing code — when tempted to clean up nearby code, add unrequested validation or options, fix something "arguably in scope", or when the diff is growing past what was asked.

A durable, disposable home for in-flight thinking that overflows a single context window — layered scratch → (optional) blackboard → (optional) conclusions under thoughts/<task-slug>/, refuted by the existing adversarial-verify and drained into durable truth (scribe / deep-analysis) when it settles. Trigger: ONLY when one task won't fit in one context (multi-session work, or a deep-work/deep-analysis run whose shared board must survive the window). Below that bar, think in-context.

8 hooks across 4 events