Self-improving, spec-driven multi-agent SDLC delivery framework — agent-agnostic plugin for Claude Code / Cursor / Copilot / Codex. v2: acceptance escrow, three-valued gates, process-health ledger, lessons library, visual-parity playbook.
Agent
Use this agent (typically via the uat-triage workflow) to validate ONE reported UAT bug - map it to requirements and specs, trace the code path, and deliver a verdict with a root-cause hypothesis. One bug per invocation.
Agent
Use this agent to implement one OpenSpec capability slice end-to-end by executing its tasks.md - schema, domain logic, services, server actions, UI - following the project's module conventions. One slice per invocation.
Agent
Use this agent (typically via the review-gate workflow) to review a diff or capability for correctness, error handling, framework best practices, and maintainability. Returns structured findings with file:line evidence.
Agent
Use this agent (typically via the eval-suite workflow) to grade ONE produced output against its rubric and return a 0-100 quality score with a pass/fail verdict. It judges quality that assertions cannot — error-message clarity, empty-state usability, copy tone — and is never the agent that produced the output…
Agent
Use this agent at phase boundaries, on demand, or at abandonment/handover to audit the GATE CHAIN (not the product) from the process-health digest, QA reports, and correction artifacts. Emits docs/qa/process-defects.json + improvement proposal folders. Fresh maker≠checker reader — MUST be dispatched as a plain…
Agent
Use this agent to author and maintain the QA proof pack (traceability matrix, manual test plan, demo script, risk register, acceptance report), handoff docs (current-state.md), technical docs, and customer-facing delivery/bugfix reports with effort logs.
Agent
Use this agent to turn raw customer input (RFP, notes, PDF, transcript) into a product brief and a fully numbered requirements document (FR/NFR/TC/BC), plus a batched clarification list. Use at project intake (Phase 1) and whenever scope changes mid-project.
Agent
Use this agent (typically via the review-gate workflow) for security review of a diff or the whole codebase - authz matrix, OWASP top 10, secrets hygiene, session handling, injection, multi-tenant data isolation. Returns structured findings with evidence.
Agent
Use this agent (typically via the review-gate workflow) to audit implemented code against its OpenSpec requirements and scenarios - finding unimplemented scenarios, silent scope drift, and behavior that contradicts the spec. Returns structured findings.
Agent
Use this agent to author OpenSpec artifacts - baseline capability specs (Phase 2) and per-slice change folders with proposal.md, design.md, and tasks.md (Phase 4a). It writes specs in strict OpenSpec requirement/scenario format that passes openspec validate --strict.
Agent
Use this agent to build and maintain the four test layers - unit (Vitest), real-DB integration, Playwright E2E, and per-slice DB smoke flows - plus deterministic seed data. Writes the slice's tests FIRST (red) from the spec, before implementation. Use in Phase 4 (tests-first, step b) per slice and Phase 5 for the…
Agent
Use this agent (typically via the vision-verify workflow) to LOOK AT a rendered UI screenshot and judge whether it visibly demonstrates a requirement and is legible — catching defects that code review and DOM assertions are blind to (dead-looking controls, poor color contrast, broken/overlapping layout, wrong-moment…