context-management

Interact with Chrome via the Claude in Chrome MCP — screenshots, DOM inspection, clicking, typing, navigation, console/network reads. Use for browser verification of UI/frontend changes. Saves screenshots to file; never returns them inline.

Execute well-defined, fully-specified code edits dispatched by the primary agent. Use for mechanical, scoped changes where the files, content, and conventions are already decided. Do NOT use when judgment, design decisions, or open questions are involved.

Look up library and framework documentation for up-to-date API references and code examples — preferably via the Context7 MCP if available, otherwise via WebFetch. Use whenever you need authoritative API signatures, SDK usage, or config details for a third-party library, instead of reading huge docs into the main context.

Review the current diff for code quality, bugs, performance, and security. Use before commit, after the implementation is complete and tests pass. Returns a structured list of findings — no files are edited.

Execute test suites (unit and/or E2E) and return a structured pass/fail report. Use whenever you need to run tests — keeps verbose test output out of the main context. Does NOT write, modify, or fix code.