AI-OS

Execute build commands, dev servers, scripts, and interpret their output. Use when: running builds, starting/stopping dev servers, executing install commands, running linters or type-checkers, interpreting non-zero exit codes, or orchestrating multi-step command sequences. Manages terminal lifecycle and output parsing.

Explore and understand an existing codebase before implementing. Use when: starting work on an unfamiliar codebase, finding where a function is defined or called, looking for existing patterns to follow, checking if a feature already exists before building it, or doing cross-repo code intelligence. Uses gitnexus MCP and workspace search tools.

Aggressive, thorough code review. Use when: reviewing a PR, auditing a file before merge, checking for security issues, OWASP Top 10 violations, performance problems, missing error handling, poor logging, or just reviewing a junior dev's work. Uses think.criticize + workspace analysis.

Systematic error triage and fix loop. Use when: build fails, type errors appear, runtime exceptions occur, get_errors returns problems, tests fail with unexpected errors, logs show exceptions. Reads VS Code diagnostics, terminal output, and log files to form hypotheses and drive fixes to zero errors.

Git operations with enforced conventions: branch from issues, conventional commits, clean history, tagging releases. Use when: starting work on a new feature or bug fix (create branch), committing completed work, creating a release tag, cleaning up branches, or enforcing commit message standards. Does NOT push or force-push without confirmation.

GitHub operations: create/update issues, open PRs, review PR status, check workflow runs, search code across repos, manage releases, list branches. Use when working with GitHub repos, tracking work in issues, or managing the release process. Uses github MCP.

Start or resume a work session. Use when: beginning a new conversation on an existing project, resuming after a break, onboarding to a codebase, or needing to establish what was done and what's next. Searches mem0 for project history, reads TODO.md, and produces a focused next-steps plan.

Pull live, up-to-date library and API documentation. Use when: asking how to use any external package, framework, or API; before implementing against a third-party SDK; when training data may be stale (anything released after 2023); looking up config options, migration guides, or changelogs. Uses context7 MCP.

Write tests, run them, fix failures, repeat until green. Use when: implementing new features (write tests first or after), a test suite is failing, need to add coverage for a bug fix, doing TDD, running E2E tests with playwright, or driving the red-green-refactor loop. Covers unit tests and E2E tests.

Structured reasoning before implementing. Use when: planning a feature, designing architecture, evaluating an approach, criticizing a draft implementation, solving a complex problem, debugging a non-obvious issue. Wraps the think MCP (think / plan / criticize).

Web research and browser automation. Use when: scraping a page, filling out a form, taking a screenshot, navigating a dynamic JS app, doing provenance-tracked web research, or when context7 doesn't cover a web resource. Uses playwright MCP for dynamic pages and context-mode for static provenance-tracked fetches.