Work planner agent. Breaks tasks into structured PLAN.md with actionable steps, file ownership, test criteria, and verified references. Produces plans developers or workers can execute without follow-up questions.
Retrospective agent. Reads spectrum artifacts (HOOK.md, debriefs, SEAM-CHECK.md, PAX-PLAN.md) and drafts LESSONS.md capturing what worked, what failed, and recommendations. user: "write up lessons from this spectrum run" assistant: uses Brown to aggregate artifacts into a structured LESSONS.md draft user: "draft the…
Lightweight parallel dispatcher. Writes SPLIT.md (file ownership), drops workers in worktrees, runs post-merge quality gate. Invoke for 3+ independent tasks or when the user says 'drop workers' / 'run parallel'.
Multi-framework test runner agent. Auto-detects pytest, jest, vitest, go test. Runs tests with coverage, generates missing tests using batch-generate-validate loops, reports failures with diagnosis.
Use this agent when a user provides a PRD, product spec, technical specification, or any structured product document and needs it decomposed into well-scoped, individually shippable Jira tickets with clear acceptance criteria. This agent should be invoked when a product manager, engineer, or team lead needs to…
Product research and problem identification agent. Validates that a problem is real, significant, and worth solving before solution exploration. Produces a structured Problem Brief grounded in evidence-based discovery frameworks. user: "what problems should we focus on for our dev tools product?" assistant: uses…
Parallel implementation agent (also called 'worker'). Executes a scoped task in an isolated worktree with strict file ownership. Spawned by Gold during parallel execution.
Post-merge spec compliance verifier. Reads PLAN.md acceptance criteria and verifies the merged codebase satisfies them. Produces SENTINEL-REPORT.md with per-criterion PASS/PARTIAL/FAIL verdicts. user: "did we build what we planned?" assistant: uses Obsidian to verify merged code against PLAN.md acceptance criteria…
Adversarial reviewer for CONTRACT.md and MANIFEST.md. Challenges Gold's decomposition before freeze — finds ownership gaps, contract ambiguities, and decomposition flaws. Runs in Phase 1.5 (The Passage) before Howlers are dropped. user: Gold has written MANIFEST.md and CONTRACT.md and wants them reviewed before…
Product strategy and management agent for PRDs, prioritization, roadmaps, and market analysis. Invoke when the user wants to think through what to build, why, and in what order — not how to build it technically. user: "help me write a PRD for the new reporting feature" assistant: uses product-strategy-partner to draft…
Behavioral spec agent. Produces DESIGN.md — a detailed API contract, schema definition, or component hierarchy spec that Howlers use as ground truth during implementation. Invoke before muster when the spectrum involves shared interfaces, database schemas, or cross-Howler API contracts. user: "we need a design spec…
Proactively remember project conventions, architecture decisions, gotchas, and corrections during normal coding work. When an agent: (1) Discovers a naming convention or code pattern, (2) Learns why something is built a certain way, (3) Encounters a non-obvious gotcha or workaround, (4) Gets corrected by the user on…
Load persistent codebase memory at session start. Automatically recalls project conventions, architecture decisions, and context from previous sessions. When an agent needs to: (1) Start a new session with full project context, (2) Recall what was learned in prior sessions, (3) Restore project memory, or (4)…