| briefer | Haiku | Read, Glob, Grep | Reads context files, returns structured briefings |
| planner | Sonnet | Read, Glob, Grep, Agent | Produces task breakdowns with behavioral test specs |
| worker | Sonnet | Full toolset | Strict TDD implementation (feature/bugfix). Red → green → regression commits. |
| worker-refactor | Sonnet | Full toolset | Behavior-preserving refactors. Existing tests must pass before and after. |
| worker-test | Sonnet | Full toolset | Adds tests to existing untested code. Mutation-checks its own tests. |
| worker-investigation | Sonnet | Read, Bash, Glob, Grep | Read-only research. Returns structured findings. |
| reviewer | Opus | Read, Bash, Glob, Grep | Code review with severity ratings. |
| ui-tester | Sonnet | Read, Bash, Glob, Grep | Visual quality inspector via browser automation. |
| ux-tester | Opus | Read, Bash, Glob, Grep | Usability evaluator via browser automation. |
| system-tester | Sonnet | Read, Bash, Glob, Grep | Full test suites, regression coverage, integration. |
| intent-validator | Opus | Read, Glob, Grep | Validates work matches user's original intent. Foreground. |
| learning-extractor | Opus | Read, Glob, Grep, Bash | Surfaces learnings from task artifacts and transcripts. |
| scribe | Haiku | Read, Write | Writes all state files. |
| issue-implementer | Opus | Agent | Autonomous groomed-backlog → tested-PR loop. Claims status:ready issues, implements each via TDD workers in dedicated worktrees, opens a PR, then loops to the next ready issue. Started via --agent issue-implementer; runs unattended after invocation. See docs/issue-implementer-runbook.md for preconditions, label setup, stop conditions, and first-run guidance. |
| issue-groomer | Opus | Agent | Autonomous backlog groomer. Claims open status-less issues, researches each deeply (codebase + product docs + external APIs), fills the target repo's issue template exhaustively with behavior contracts, assumptions, and alternatives, then applies status:ready (clear implementation path) or status:blocked (exhaustively groomed, genuine product/path decision needed). Loops; in WATCH mode re-checks every poll_interval for new issues. Started via --agent issue-groomer; runs unattended after invocation. See docs/issue-groomer-runbook.md for preconditions, label setup, stop conditions, and first-run guidance. |
| researcher | Sonnet | Read, Bash, Glob, Grep, WebFetch, WebSearch, Write | Deep external research on libraries, frameworks, SDKs, APIs, CLI tools, or cloud services. Reads official docs, source code, examples, runtime probes. Returns structured findings — capability map, mental model, failure modes, expectation gaps. No code commits. |
| distiller | Opus | Read, Glob, Grep, Write | Compresses raw evidence (research notes, command/error logs, debug transcripts, code samples) into citable knowledge artifacts: gotchas, patterns, anti-patterns, playbooks, before-you-build guides, decision records. Every claim must cite evidence. Returns a distillation manifest. |
| knowledge-packager | Sonnet | Read, Glob, Grep, Write | Assembles distilled knowledge into a navigable, self-contained artifact (skill pack, runbook, onboarding guide, internal wiki). Writes a cross-linked directory with index, quickstart, lessons, labs, recipes, checklists, troubleshooting, and agent instructions. Returns a navigability-verified manifest. |
| evidence-auditor | Opus | Read, Bash, Glob, Grep | Read-only auditor of claim-vs-evidence integrity. Verifies documented claims have supporting evidence; evaluates named quality gates (research-quality, distillation-quality, skill-pack-quality). Returns per-gate verdicts and unverified-claim findings with severity ratings. No file modifications. |