Subagents
Copyable Claude Code subagent definitions with tuned system prompts, tool allowlists, and model routing. Save one under .claude/agents and Claude Code spawns it on demand or when its description matches the work.
14 definitions
Reviews a diff for correctness bugs and reuse or simplification cleanups, ranked by confidence.
Drafts decision-intent comparison content that is sourced, balanced, and free of hype. Amusingly literal about the no-em-dashes rule, but it holds.
Plans and executes a database migration behind a dry run, with a written rollback before anything mutates.
Writes tests that pin real behavior and would actually fail on the bug, matching the repo's existing test style.
Drives the running app in a browser, captures evidence, and reports UX and design-contract breaks with screenshots.
Fact-checks claims against primary sources and flags anything unsupported, stale, or misattributed.
Turns merged commits into a clean, user-facing changelog grouped by change type, dropping the noise.
Read-only survey that finds duplication, dead code, and the highest-leverage cleanups, then ranks them by payoff.
Documents endpoints and functions from the code itself, with real request and response examples that match the implementation.
Works a production incident from symptom to root cause using logs and the code, and proposes the smallest safe fix.
Maps the content graph and proposes contextual internal links that pass real relevance, never a footer link dump.
Drives the app at real mobile viewports, hunts horizontal overflow and clipped content, and reports each break with a screenshot.
Reads a failing Vercel build log and pins the real cause: blocked build scripts, the type gate, lockfile drift, or a missing env var.
Verifies every checkable claim in a draft against a primary source before it publishes, and blocks the ones it cannot source.