mblauberg

academic-writing

Use for drafting, rewriting, condensing, or checking academic prose and LaTeX while preserving claims and citations. Not for general, engineering, or Australian legal prose; use the matching writing skill.

autopilot

Use for a large, open-ended mission that must survive sessions/crashes and run user-out-of-loop until STOP; delegates lifecycle work to other skills. Not for pre-scoped delivery that stops at a PR gate; use orchestrate's autonomous mode or deliver.

Code & Development Listed

code-review

Use for source-read-only review of a PR, diff, commit, migration, refactor, or implementation through its dependency cone. Not for root-cause diagnosis, authorised fixes, or browser/UX review; use diagnose, implement, or ui-ux-design.

d2-diagrams

Use for creating, validating, rendering, or exporting D2 source, including fixed-layout and vector-PDF publication figures. Not for Mermaid, Graphviz, or PlantUML requirements diagrams; use engineering-docs or uml-diagrams.

deliver

Use for taking an approved research, analysis, document, or agent-product outcome through evidence, review, and user acceptance. Not for software-only work, unsettled scope, or release; use implement, scope, or release.

diagnose

Use to reproduce and find the root cause of broken behaviour, crashes, test failures, flakes, or performance regressions. Not for a known-cause repair or change review; use implement or code-review.

engineering-docs

Use for creating, placing, indexing, updating, or archiving engineering docs and docs structures. Not for prose-only rewriting or session continuity; use engineering-writing or session.

engineering-writing

Use for drafting or condensing software-engineering prose such as requirements/specs, READMEs, commits, PRs, runbooks, errors, and design notes. Not for general, academic, or Australian legal prose.

evaluate

Use for repeatable quality or safety evaluation of stochastic or judgement-bearing systems, prompts, agents, rankings, or artifacts. Not for deterministic tests or ordinary code review; use tdd or code-review.

Code & Development Listed

github-setup

Use when the user says "set up GitHub for this project" or asks to scaffold a new repo's labels, branch ruleset, issue forms, PR template, work-item runbook, or Project board. Not for this repo's own day-to-day GitHub mechanics; use its runbook.

implement

Use for an approved software change through verified implementation, independent review, bounded repair, and user acceptance. Not for unsettled scope, diagnosis-only, read-only review, or release.

legal-writing

Use for drafting, reviewing, condensing, or source-checking Australian legal documents (forms, affidavits, submissions and orders), legal decision overviews, or correspondence. Not for academic, engineering, or general prose; use its specialist.

natural-writing

Use for making general or mixed prose natural and direct while preserving facts, citations, numbers, and intent. Not for engineering, academic, or Australian legal prose; use its specialist.

orchestrate

Use when bounded fan-out, multi-agent research, cross-family review, parallel audits, Herdr control, or autonomous ready-issue implementation helps. Not for tiny work, coupled debugging, or run-until-STOP jobs; use diagnose or autopilot.

playwright

Use for terminal-driven browser navigation, interaction, snapshots, screenshots, extraction, or UI-flow debugging. Not for authoring Playwright test suites or frontend design; use tdd or ui-ux-design.

prototype

Use for an authorised, timeboxed throwaway build that answers one feasibility question. Not for production code, settled requirements, or debugging known breakage; use implement, scope, or diagnose.

Web & Frontend Listed

react-performance

Use for measuring or improving React rendering, hydration, bundles, waterfalls, Server Components, or Core Web Vitals. Not for general UI design, TanStack cache semantics, or Vite config; use the matching specialist.

Code & Development Listed

refactor

Use for authorised, behaviour-preserving structural simplification backed by characterisation and equivalence evidence. Not for new behaviour, read-only review, or unknown root cause; use tdd, code-review, or diagnose.

release

Use for user-authorised promotion of an accepted artifact: deploy, publish, send, roll out, or observe. Not for implementation or drafting; use implement or the domain owner.

retrospect

Use after delivery, release, incident, evaluation, or a long run to derive evidence-backed process improvements and regression gates. Not for session cleanup, one skill audit, or an active defect; use session, skill-craft, or implement.

scope

Use when unsettled requirements or decisions need a spec, options, acceptance criteria, authority bounds, or decision record. Not for implementing approved scope or writing files during read-only advice; use implement.

session

Use for start, checkpoint, handoff, compaction, or end-of-session continuity on substantial work. Not for a multi-session route map or read-only write authority; use work-map for effort state.

skill-craft

Use for creating, revising, or read-only auditing an Agent Skill: SKILL.md, triggers, progressive disclosure, fixtures, overlap, and token cost. Defaults to audit; edits need explicit authority. Not for plugin packaging or delivery; use implement.

API & Backend Listed

tanstack-query

Use for TanStack Query v5 server-state code: keys, options, freshness, mutations, invalidation, optimistic updates, pagination, cancellation, or SSR. Not for generic React state, performance, routers, or backend caches.

tdd

Use for writing the first right-reason failing test for new or changed observable behaviour, then red-green-refactor. Not for behaviour-preserving structure, diagnosis-only, or delivery; use refactor, diagnose, or implement.

typescript-clean-code

Use as a TypeScript or JavaScript correctness lens for type modelling, strictness, null/error boundaries, async ownership, and tests. Not a lifecycle owner or reason to rewrite stable conventions; combine with the task owner.

Web & Frontend Listed

ui-ux-design

Use for frontend UI/UX: read-only review (default — visual, accessibility, responsive, interaction) or authorised design, redesign, and live iteration. Not for source-diff review or React profiling; use code-review or react-performance.

uml-diagrams

Use for requirements-spec UML in PlantUML: use-case, package, activity, include/extend, and swimlane diagrams. Not for D2, Mermaid, C4, Structurizr, or unrelated PlantUML; use the matching diagram owner.

work-map

Use when a multi-session effort needs one durable, curated route linking its specification and work items. Not for live status, ownership, dependencies, user gates or session handoffs; use the project's work tracker and session.

setup-repo

Use when asked to set up or scaffold a repository's process, labels, tracker, branch ruleset, issue forms, docs layout, work-item runbook, or board. Not for this repo's own day-to-day GitHub mechanics; use its runbook.

caveman

Use when the user invokes Caveman or requests aggressively compressed chat. Full and ultra remain user-explicit. Not for generic brevity, a governing terse default, or artifact condensation.

grill-me

Use when the user asks to be grilled or wants one-question-at-a-time stress testing of a plan or design. Not for writing the final spec or implementing it; use scope or implement.

Web & Frontend Listed

web-stack-conventions

Use for current deltas in Vite 8/Rolldown/Oxc, WCAG 2.2, Lighthouse 13, or OpenAPI 3.1. Not for full frontend design, React performance, or generic API implementation; use the relevant owner.