Claude Skills by instrument-org

Readable upstream source on demand by name, via the agent-reference CLI. Use when a task needs a library's real source rather than a memory of it, so writing code against an API you cannot recall exactly ("use the combobox from this component library", "wire this up with X"), or asking how X implements something, how its maintainers test it, why it behaves this way, or whether it is worth adopting. Use it before reading a dependency's published build to answer a question about it, anything un...

Approach

Find recent product UI surfaces changed by Instrument commits that may need human review. Use when asked to find UI changes, surface UI changes, create a design or UI review queue, prepare a designer handoff, or identify app changes needing design attention.

Generate a git commit message matching the Instrument monorepo's scope-first commit style. Use when the user asks for a commit message, wants to commit changes, or asks how to describe their changes. Knows the repo's scopes (studio, workspace, dx, etc.) and real examples from the commit history.

Build a static HTML wireframe of a proposed Studio flow, styled with the app's real design tokens and Tailwind. Use when a plan or design brief needs a picture, when asked for a wireframe or mockup of a feature, or when a proposal would be clearer as a sequence of UI states.

Generate stoic, human-readable release notes grouped by product area for a general audience. Use when asked for release notes since a version.

Test bash commands in the same just-bash sandbox the agent uses, without booting Studio. Use when validating bash environment fixes, checking command availability, or verifying tool behavior (uv, pnpm, tsx, ffmpeg, etc.).

Export a task's session as a markdown transcript from a task directory or an exported task .zip. Use when asked to dump, export, or read a session/task as markdown, review an agent run offline, or inspect the transcript inside a shared task zip.

Use when working with Chrome DevTools against the Studio Electron app in this repo. Load this for Studio-specific connection details and page selection. Especially relevant when the user mentions Electron, `chrome-devtools`, `chrome-devtools-cli`, browser views, smoke tests, or debug routes.

Read Studio main-process logs written during local development. Use when you want to inspect what happened during a Studio session — errors, warnings, or info messages from the Electron main process.

Read task .instrument/task.db files with safe read-only SQL. Use when an Instrument task investigation needs raw messages, parts, sessions, or any other stored records.

Start, inspect, stop, and drive an installed Instrument Studio build on a preconfigured Linux desktop host through SSH and loopback-only CDP. Use for Linux-specific Electron validation, remote Studio smoke tests, installed-product checks, or any request to test Studio on a Linux machine from another host.

Start, inspect, stop, and drive Instrument Studio development or installed builds on a preconfigured Windows host through SSH and loopback-only CDP. Use for Windows-specific Electron validation, remote Studio smoke tests, installed-product checks, or any request to test Studio on a Windows machine from another host.

Use the `typescript-result` library for type-safe error handling in TypeScript. Replaces try/catch with Result<T, E> and AsyncResult<T, E> types, enabling compile-time error tracking, chaining, and generator-style composition. Use when writing functions that can fail, wrapping async operations, handling errors exhaustively, or when the user mentions Result types, error handling patterns, or typescript-result.

Choose how to check an Instrument change — sandbox shell, tool harness, a real agent across models, or the running app. Use after changing an agent tool, a prompt, a skill, the sandbox, or any behavior reading the code cannot confirm.

Create a tailored, single-file HTML visual explanation of a complex question, code change, architecture, plan, incident, comparison, workflow, or product behavior. Use when the user asks to visualize work, make something easier to understand, explain changes visually, produce an interactive technical artifact, or invokes this skill for a richer answer than prose alone.

Create or revise a reusable agent skill from a user's workflow, domain knowledge, or tool instructions. Use whenever a user wants to create, capture, package, edit, or turn something into a skill.