assistant

iOS code signing and provisioning profile management via fastlane match. Use when debugging CI signing failures, adding new app extensions or targets, rotating certificates, or troubleshooting match decryption errors.

Enforces OpenAPI spec discipline when working on REST API endpoints in this project. Triggers whenever adding, modifying, or removing HTTP routes, request/response types, or API handlers in the Rust web-ui crate (`crates/web-ui`). Reminds the agent to (1) update the committed `openapi.json` spec, (2) run `make dump-openapi` to re-export the spec from the running server, and (3) run `make generate-flutter-client` to regenerate the Dart/dio client in `app/packages/assistant_api/`. Also applies when changing route parameters, status codes, or authentication on existing endpoints.

Test Driven Development discipline for this workspace. Use whenever writing new features, fixing bugs, or refactoring — the test must be written and confirmed failing before any implementation code is added.

CI organization principles and workflow structure for this project. Use whenever modifying, adding, or reviewing GitHub Actions workflows — including adding jobs, changing job order, updating action versions, modifying cache keys, adding build targets, or debugging CI failures.

Playwright visual regression testing for the assistant Flutter SPA. Covers test structure, screenshot baselines, cross-platform diff tolerance, CI workflow with inline diff comments, and baseline management. Use when adding screens, changing layouts, or debugging visual test failures.

Checklist and architectural rules for implementing new assistant interfaces (CLI, Slack, web, etc.). Every interface MUST use the Orchestrator to ensure consistent behaviour: system prompt, tools, skills, memory, and the full ReAct loop. Use this skill when adding, reviewing, or fixing an interface.

Flutter UX/UI principles for the assistant app. Covers dark-theme colour tokens, responsive breakpoints (Flutter adaptive layouts), accessibility, Riverpod state patterns for loading/error/empty states, and platform conventions for web and macOS. Use when building or reviewing any Flutter screen or widget in app/.

Rust dependency management principles for this workspace. Use whenever adding, updating, removing, or reviewing dependencies in any Cargo.toml — including deciding which version to use, where to declare a dep, or auditing for duplicates.

Run tasks on the local machine using Claude Code CLI as a background agent. Use this when the user wants to execute code, edit files, run shell commands, build projects, analyse repositories, or do any agentic work on the device — especially multi-step tasks that benefit from Claude Code's tool-use loop. Supports fire-and-forget async jobs (non-blocking, via the native process tool) as well as quick blocking one-shot tasks and follow-up questions in the same session.

Authoritative specification for the Agent Skills (agentskills.io) SKILL.md format. Use this skill when generating or validating SKILL.md files.

Practical observability guide for this workspace. Includes where traces, logs, and metrics are persisted in SQLite and quick query patterns for debugging. Use when investigating runtime behavior, incidents, or telemetry quality.

OpenTelemetry semantic conventions for Generative AI attributes. Use when recording telemetry for GenAI.

Best-practice guide for instrumenting Rust apps with the OpenTelemetry SDK. Use when setting up traces, metrics, logs, exporters, and lifecycle handling. For GenAI attributes and semantics, pair with the opentelemetry-genai-spec skill.

Cleanup the current architecture. Look for opportunities to extract traits to make things smaller, dedupe, remove legacy or make things more efficient or testable.

Browser automation via @playwright/mcp (Microsoft). Use this when the user wants to navigate websites, fill forms, take screenshots, scrape web content, test web apps, or run any multi-step browser workflow. Requires no display (headless mode supported).

Run coding agents (Claude Code, Codex, OpenCode, or others) as background processes for programmatic control. Use when you need non-blocking execution, parallel agents, PR reviews, or long-running coding tasks. Prefer this over direct bash for any task that takes more than ~20 seconds.

A minimal example WASM skill that returns a greeting. Use to verify that the WASM execution tier is working correctly.