Runs this session as a child of a ticket's manager, the bottom of three levels. It does one named part, usually in another repo, asks and reports only to its manager, and leaves closing to it. Use when a slot opens with /orchestrator:child.
Runs this session as the manager of one ticket, the middle of three levels. It reads the ticket, starts the matching dev workflow, owns the work to release, and runs child sessions for other repos. Use when a window opens with /orchestrator:manager.
Runs this session as a product's triage, the top of three levels. It owns every ticket, the plan, status and overlap checks, starts ticket windows, and writes no code. Use when the user runs /orchestrator:triage to coordinate parallel ticket work.
Runs models through the claudish MCP tools — team, create_session, run_prompt — and resolves model IDs against the live catalog. Use when the user mentions claudish, OpenRouter, or external AI models.
Runs a review across several AI models in parallel and aggregates their verdicts, the judge shape of /multimodel:team. Use when the user wants a second opinion, multi-expert validation, or consensus from Grok, Gemini, GPT, or Kimi.
Slot-addressed tool patterns for interactive terminal access. Use when running interactive commands, starting dev servers, watching tests, querying databases, or working beside the user.
Orchestrates tmux workspaces — dashboards, watch/entr monitors, one command on many hosts — on numbered helper slots. Use for a multi-pane dashboard, an ambient monitor, a command across hosts, or a startup script for the user.
Hands a task to Aside, the user's own browser and agent, through the aside CLI. Use when the work is in their accounts (my Slack, Gmail, Linear, GitHub, calendar) or they mention Aside; use browser-use for scraping and tests.
Authors, runs and debugs madbench benches natively — bench YAML, checks, metrics with module/, red-state testdata, the two controls, exit codes; drafts an upstream issue instead of a wrapper. Use when writing, reviewing or debugging any bench.
Creates, runs, debugs and migrates madbench benches. Checks the installed madbench against these skills first, then routes the request. Use when the user types /madbench or asks to create, run, fix or migrate a bench.
Brings a madbench Bench or Experiment written for madbench v0.41.0 or later to the installed madbench, applying the changes `changes.md` records between the two versions. Use when a bench no longer loads after an upgrade.
Runs a multi-model review gate to a verdict: start the claudish team panel with input_file only, poll until settled, read every ballot, apply the minimum ballot count, log every skip. Use at any plan-review, code-review, verify or fix gate.
Merges independent reviews of one target into one report with a consensus tag per finding and a verdict from supplied thresholds; also synthesises research findings. Use when several reviews or votes must become one file a gate reads.
Creates, uses and cleans up git worktrees with safety checks. Use before isolated, risky or parallel feature work, or on mention of worktree, experiment or prototype.
Handle errors, timeouts, and failures in multi-agent workflows. Use when dealing with external model timeouts, API failures, partial success, user cancellation, or graceful degradation.
Sets up code-search in a project (search engine, index and settings) through its setup CLI, asking before every change. Use when code-search setup is incomplete, or the user asks to set up, reconfigure or upgrade code search.
Root-cause debugging — reproduce, localize, explain, verify — with depth routing and a technique catalogue for stack traces, wolf fence and data-flow tracing. Use when anything errors, crashes or misbehaves, even on a bare pasted trace.
Release one or more Magus plugins. Infers the version bump from git history, bumps plugin.json and marketplace.json, commits, and after the PR merges tags the merge commit with one explicit ref per tag; CI gates the PR and publishes the dist repos. Use…
Reference for the Browser Use MCP tools — parameters, returns, session lifecycle. Use when navigating, clicking, typing, evaluating JS, or extracting page content.
Combines Browser Use (DOM, clicks, screenshots) with claude-in-chrome (console, network, JS) for full-spectrum debugging. Use when investigating console errors or network requests.
Router for the eight Bun/TypeScript skills — setup, HTTP, errors, testing, security, production, performance, TUI. Says which to read for a task, without loading them. Any Bun or TS work.
Diagnose and fix Bun/TypeScript performance — event-loop blocking, measured native fast paths, benchmarking that does not lie, streaming, caching, workers, memory. Ships a tested benchmark harness.
Ship a Bun service — graceful shutdown ordering, structured logging, liveness vs readiness, multi-stage Docker, signals and PID 1, frozen lockfiles, CI. Ships tested logger, shutdown and health code plus a Dockerfile.
Start or restructure a Bun/TypeScript project — folder structure by business component, strict tsconfig, typed env config, workspaces and catalogs, bunfig, linting. Ships a tested config parser and measured install-layout facts.
Build, review, or debug OpenTUI terminal UIs — React JSX or core constructs, Yoga flexbox, gradient meters, colour screenshots. Any Bun/TypeScript TUI. Only Bun is tested; Node/Deno at own risk.
Detects the browser-use@magus plugin and captures screenshots of URL-based references. Use when designer agents need browser screenshot capture and claude-in-chrome is unavailable.
Provides design system references — Material 3, Apple HIG, Tailwind UI, Ant Design, Shadcn/ui. Use when reviewing UI against an established design system.
Branches Neon, Turso, or Supabase per git worktree for isolated schema work. Use when a worktree changes the schema, or on mention of Neon, Turso, Prisma.
Detects the project stack from its config files, then names the one or two dev skill files to read for this task — it loads none of them. Use at the start of any coding task, or on an unfamiliar or multi-stack repo.
Testing pyramid, AAA structure, test doubles, fixtures, assertions and coverage targets, in any language. Use when writing, reviewing or setting up tests. Do not use for the write-test-first loop — that is test-driven-development.
15 ranked documentation practices, 7 templates and anti-slop writing rules. Use when writing or reviewing a README, API reference, guide or changelog, even if the user only says "document this".
Generates and reviews Dingo meta-language code with Go boundary handling. Use when the user mentions Dingo, .dingo files, or Dingo-to-Go transpilation.
Provides pass/fail/running/idle output markers for 15+ test, build, and deploy tools. Use when monitoring terminal output for jest, vitest, pytest, go test, cargo, webpack, vite, or vercel.
Drives a Red-Green-Refactor state machine for TDD with terminal test watchers. Use when running TDD in watch mode or iterating on failing tests with jest, vitest, cargo watch, or pytest-watch.