A quality harness for Claude Code. Routes any prompt into spec → plan → implement → adversarial audit → living docs, and holds the turn open until the work actually holds up: no TODOs, no deferred scope, no unverified test claims.
Agent
Accessibility (WCAG) auditor for UI-touching changes. Invoke during review whenever a change adds or modifies markup, templates, components, styles, or client-side view logic: verifies semantic structure, keyboard operability, focus management, contrast, forms, ARIA correctness, media alternatives, and motion safety.…
Agent
Adversarial security and robustness auditor. Invoke during review to actively try to break a change: security vulnerabilities, injection, unsafe states, auth/authz gaps, unvalidated input, resource exhaustion, and abuse cases. Read-only; produces an attacker's-eye assessment.
Agent
CLAUDE.md regression verifier. Invoke before writing any change to a CLAUDE.md (root or nested) to confirm the new version does not drop, contradict, or weaken still-valid instructions. Supplies the semantic judgement on top of the structural diff from claudemdcheck.py. Read-only.
Agent
Completeness and scope-fidelity auditor. Invoke during review to prove the change is actually finished: no placeholders/TODOs/stubs, no NotImplemented, no debug leftovers, no dead/commented-out code, and, critically, nothing silently dropped or narrowed relative to the agreed spec. Confirms every acceptance criterion…
Agent
Technical-debt auditor. Invoke during review to judge what a change will cost later: the shortcuts it takes, the coupling and duplication-by-obligation it creates, the deprecated or pinned dependencies it leans on, the tests that lock in implementation rather than behaviour, and whether the debt it does take on is…
Agent
Design-system, UX-consistency and craft auditor for UI-touching changes. Invoke during review whenever a change adds or modifies user-facing interface: verifies design-token adherence (no magic values), typography/spacing-scale discipline, component reuse over one-off variants, state completeness, responsive coverage…
Agent
Documentation-first auditor. Invoke before or during review to confirm a change follows the authoritative documentation for the libraries/APIs it uses and matches existing in-repo patterns, and to catch reinvented wheels. Read-only.
Agent
Duplication, reinvention, and over-engineering auditor. Invoke during review to find logic the change duplicates (verbatim or near), existing utilities/functions/modules that should have been reused, and code the change adds that isn't needed: speculative abstractions, unused parameters/exports, and needless…
Agent
Edge-case and correctness auditor. Invoke during review to enumerate boundary conditions, null/empty/huge inputs, concurrency, error paths, and unusual-but-real use cases, and to verify each is handled correctly. Attends to the small things (off-by-one, types, messages). Read-only.
Agent
Reverse auditor. Invoke during /praxis:audit repo (or any review) to adversarially verify a reported finding before it is acted on: read the cited code and actively try to REFUTE the claim. Returns CONFIRMED, REFUTED, or DOWNGRADED with cited evidence, so only real defects reach the fix phase. Read-only.
Agent
Performance and scalability auditor. Invoke during review to assess algorithmic complexity, hot paths, N+1 queries, allocations, I/O patterns, caching, and how the change behaves as data volume and load grow. Read-only.
Agent
Regression auditor. Invoke during review to find behaviours, contracts, and tests the change may have broken: changed function signatures, altered return values, side effects, affected callers, and missing/failing test coverage. Read-only analysis (does not run destructive commands).
Agent
Read-only codebase mapper. Invoke during bootstrap/onboarding to understand an unfamiliar repo: its purpose, architecture, real build/test/run commands, conventions, subsystem boundaries, and integration points. Produces the map that CLAUDE.md generation is built from.