Write, change, organize, or review non-generated code; evaluate whether a maintainer can understand and safely change it; or decide source order, naming, control flow, locality, physical organization, or whether a boundary reduces cognitive and navigation…
Write, change, review, test, debug, or explain Go code; work with Go modules, tool dependencies, or dependency source; or choose Go language and standard-library APIs. Do not use when Go is incidental and no Go-specific decision, artifact, or diagnosis is…
Use when writing, changing, reviewing, testing, debugging, or explaining Rust code; designing Rust APIs and representations; or choosing Rust language and library idioms. Do not use when Rust is incidental context and no Rust-specific decision, artifact, or…
Create, revise, review, or consult Architecture Decision Records (ADRs) and decision logs, or decide whether a technical choice needs durable rationale. Do not use merely for architecture work, system descriptions, or implementation plans that do not record a…
Use when the user asks to explain or teach a concept with a Manim animation. Do not use for static diagrams, decorative motion, interactive visualizations, or editing existing footage.
Design or change command-line interfaces: commands, flags, output streams, structured output, exit status, and the boundary between command adapters and application behavior. Do not use merely to invoke an existing command.
Use for writing, revising, reviewing, or deciding whether to add or remove source comments and documentation; introducing or changing domain concepts, non-obvious contracts, invariants, or representation boundaries; or unclear local code models. Not review…
Design, change, or evaluate code ownership, boundaries, contracts, representations, dependency lifetimes, domain adapters, public API shapes, or where policy and state live, including new code and refactoring. Do not use for a local readability-only edit that…
Decide what test evidence a code change needs; write, change, delete, or review tests and their supporting infrastructure. Do not use merely to explain an existing test result or compiler error when no code, test, or validation decision is requested.
Draft, revise, evaluate, review, or apply commit messages; commit, amend, create fixups or branches for commits, stack work, insert prerequisites below the current branch, or recover after raw Git. Do not use for general prose or code review involving neither…
Plan, create, revise, or review an explanatory 2D diagram's viewer task, visual argument, visual vocabulary, spatial layout, or semantic correctness, including when another diagram skill requires it. Do not use for quantitative charts, decorative…
Use for Git commits, amendments, fixups, branch creation or movement, stacks, pushes, pull request creation or metadata, merges, review submission, publication, or recovery from raw Git. git-spice is mandatory for supported repository operations.
Use when the agent is responsible for reviewing a diff, commit, pull request, or code change and producing behavior, design, readability, documentation, or reviewability findings. Excludes responding to feedback on its own changes and helping the user conduct…
Use when the user explicitly requests a Pikchr diagram, Pikchr source, a `.pikchr` file, or SVG rendered with Pikchr. Do not use for a general diagram request that does not select Pikchr as the format.
Use when a diagram must be readable as plain text, including when the user requests text characters or the destination is a plain-text artifact such as Markdown, a commit message, or terminal output. Plain-text diagrams can also be used in chat when they suit…
Use when writing or editing prose other than ordinary conversational chat, including formatting-only edits, or when the user requests source-style prose in chat.
Use for substantive explanations, comparisons, proposals, and multi-part answers in conversational chat; or when writing or substantially revising reader-facing prose outside the current conversation, including documentation, design documents, incident…
Use when receiving or addressing code review feedback about changes you made, including pasted reviewer comments, inline PR comments, requested changes, review summaries, questions, nits, or direct user critique that may require replies or revisions.
Use when creating, editing, validating, rendering, or debugging local CLI and TUI demo GIFs with asciinema+agg, Betamax, or VHS, including recorder selection, scripted terminal recordings, `.tape` files, captions, keyboard overlays, waits, pacing, and…
Use when the user requests a subagent quorum, independent design proposals, panel-based synthesis, or a multi-agent tie-break between defined alternatives. Do not use for ordinary delegation, parallel fact gathering, implementation splitting, or a choice that…
Use when acquiring, using, inspecting, or returning temporary pooled Git worktrees with the `treehouse` CLI for isolated task work, delegated subagent work, or cleanup of leased treehouse worktree paths.
Use when drafting, revising, reviewing, or evaluating a commit message, pull request title or description, or source text intended to become or supply a commit message or pull request metadata. Do not use for ordinary change explanations, changelogs, release…
Create, revise, or test agent instructions in skills, AGENTS.md, CLAUDE.md, and their references. Use for instruction routing and behavioral evaluations, not ordinary prose.
Use when creating pull requests, updating pull request branches, pushing or publishing work for review, applying pull request metadata, or submitting stacked branches. Do not use for drafting or reviewing pull request metadata without a pull request…