poshan0126

dotclaude

Reviews code for real bugs: off-by-ones, null derefs, race conditions, swallowed errors. Skips style nitpicks.

OWASP-style review: injection, auth, data exposure, weak crypto. Severity-ranked with file:line + fix.

Finds real bottlenecks. DB N+1, memory leaks, blocking calls, re-renders.

Cross-references docs against actual source. Flags drift, missing params, dead references.

Builds production-grade UI. Tokens-first, anti-AI-slop aesthetics, design tokens enforced.

Finds code that fails silently: swallowed errors, failures masked as success, fallbacks that hide breakage. Runs on every PR review.

Judges whether a diff's tests actually verify the change: assertion-free tests, mock theater, tests that cannot fail, weakened tests.

Deterministic guardrails as PreToolUse hooks: blocks dangerous commands (force push, rm -rf, DROP TABLE), secret leaks, edits to protected files, and writes to build artifacts.

Set up dotclaude in any project. Deep-scans the codebase, interviews you, then installs and customizes only the components the project actually needs.

Find and fix a bug. Default is the careful path (reproduce, investigate, test). Add `--fast` for emergency hotfix mode.

Commit, push, create PR with confirmation at every step, then optional cleanup of branches whose remotes are gone. Blocks secrets and force-push.

Review PR or staged changes via six specialist agents in parallel. Confidence-bucketed, severity-ranked synthesis.

Strict red-green-refactor TDD loop. Failing test, then minimum code, then refactor. Commits after each cycle.

Explain code with one-sentence summary, mental model, ASCII diagram, and modification guide.

Safely refactor with tests as a safety net. Writes tests first if none exist. Use --diff to simplify just the working diff before committing.

Write comprehensive tests covering happy/edge/error/concurrency paths. Verifies tests actually catch bugs.

Estimate per-turn token cost of your .claude/ configuration and CLAUDE.md. Always-loaded vs path-scoped vs invoked-only, with top contributors. Use --api for exact counts.

Rebuild context fast after /clear: reads the handoff note and branch changes, summarizes where work stands. Add handoff to write the note before stopping.

Keep CLAUDE.md current and lean: capture session learnings into it (default) or audit it for stale commands, drift, and bloat.

Take a GitHub issue from number to tested fix: read the thread, reproduce, fix with a regression test, prep a PR that closes it.