Agents
20 agents — 15 independent review critics and 5 orchestrating leads/gates. This roster is generated from the real agent frontmatter at build time.
Review critics critic
Independent and read-only. They audit diffs inside code-review and never edit code or approve their own findings.
| Agent | Role |
|---|---|
accessibility-advocate critic | Independent accessibility critic (WCAG 2.1/2.2 AA). Reviews UI changes for semantics, keyboard operability, focus management, name/role/value, contrast, and assistive-technology support, and flags violations with the WCAG criterion and a fix. |
architecture-advocate critic | Independent architecture critic. Reviews a change for module boundaries, coupling/cohesion, state ownership, API/type contracts, and dependency direction, and flags designs that will be costly to change. |
code-reviewer critic | Independent correctness-and-quality critic for a code change. Reviews a diff/PR for logic errors, edge cases, error handling, readability, and regression risk, and returns severity-tagged findings. Dispatched by review/code-review. |
debugger critic | Reproduce-and-isolate specialist. Given a failure, reproduces it deterministically, narrows to the smallest failing case, and localizes the root cause with evidence — pairs with analyze-bug and root-cause. |
dependency-manager critic | Independent dependency critic. Reviews added/updated dependencies for necessity, security advisories, license compatibility, maintenance health, and supply-chain risk, and flags anything that shouldn't be pulled in. |
documentation-steward critic | Independent documentation critic. Reviews whether a change keeps docs, READMEs, public API references, and comments truthful and in sync, and flags drift, missing docs for public contracts, and misleading comments. |
e2e-test-author critic | Independent end-to-end test critic and author-advisor. Reviews whether critical user journeys are covered by resilient Playwright E2E tests (stable selectors, real assertions, flake resistance) and recommends the journeys worth automating. |
internationalization-expert critic | Independent i18n/l10n critic. Reviews changes for hardcoded strings, formatting (dates/numbers/currency), pluralization, RTL support, and locale-safe logic, and flags anything that breaks translation or non-English locales. |
manual-tester critic | Independent exploratory-testing critic. Probes a change for boundary conditions, error states, unusual inputs, and race/interaction bugs that automated tests miss, and reports concrete, reproducible issues. |
observability-advocate critic | Independent observability critic. Reviews whether a change is diagnosable in production — structured logs, metrics, traces, error reporting, and privacy-safe signals — and flags blind spots and PII leaks. |
performance-advocate critic | Independent performance critic. Reviews a change for algorithmic complexity, N+1 and chatty I/O, allocation/memory, bundle size, rendering, and caching, and flags regressions with evidence. |
refactorer critic | Independent maintainability critic. Reviews a change for duplication, over-long functions, unclear naming, and abstraction thresholds, and recommends focused refactors — without demanding speculative generality. |
repair-validator critic | Independent gate for autonomous or assisted repairs. Verifies that a proposed fix actually resolves the failure it targets, introduces no regression, respects scope, and meets quality gates — before any human merge. Read-only and adversarial; it does not author fixes. |
security-advocate critic | Independent security critic (EASE-MAS A9). Reviews a change for authn/authz, injection, secrets, unsafe deserialization, SSRF, insecure defaults, and supply-chain risk, and returns severity-tagged findings with concrete fixes. |
style-guardian critic | Independent UI/visual-consistency critic. Reviews changes for design-token usage, responsive behavior, spacing/typography consistency, motion, and visual polish — distinct from accessibility and code style. |
unit-testing-advocate critic | Independent unit-test critic. Reviews whether a change is covered by meaningful unit tests — contracts, equivalence classes, boundaries, and error paths — not just line coverage, and flags weak or missing tests. |
Leads & gates lead
Orchestrators and validators that drive multi-stage flows and guard merges.
| Agent | Role |
|---|---|
a11y-tester lead | Accessibility audit lead. Runs the full accessibility lifecycle for a component or page — scan, deep review, remediation, verification, and regression-test generation — and produces a consolidated WCAG 2.1/2.2 AA report. Orchestrates the accessibility skills. |
bug-exterminator lead | Bug-fix lead. Drives a defect from reproduction through root-cause analysis to a bounded, verified repair — reproduce → analyze → hypothesize → fix → verify → regression-gate — and hands off to repair-validator before any human merge. Enforces repair bounds. |
supervisor lead | Closed-loop orchestrator. Drives the EASE-MAS lifecycle (requirements → PRD → architecture → code → test → execution → failure → RCA → repair → re-validate) as a DAG, enforces autonomy/permission levels and human-in-the-loop gates, and delegates to leads/critics/skills. Never merges autonomously. |
traceability-keeper lead | Maintains the traceability graph that links requirements → PRD items → architecture decisions → code symbols → tests → executions → failures → bugs → hypotheses → repairs → validations. Records nodes and edges as work progresses so impact and coverage are always queryable. |