Agent
Use this agent for Phase 1 investigation when the task requires codebase exploration — systematically catalogs the current state of a concern. Searches for patterns, construction sites, and edge cases. Returns a structured findings report.
Agent
Use this agent for Phase 1 investigation when the task requires both codebase exploration AND external knowledge — e.g., understanding current implementation then researching how an API or library actually works. Returns a unified findings report.
Agent
Use this agent for Phase 6 code quality evaluation — reviews implementation for code quality, security, type safety, or other concerns. Dispatched with a specific evaluator prompt from the project's evaluator configuration. Runs AFTER spec compliance passes.
Agent
Use this agent to address specific findings from evaluators or spec-reviewers that indicate spec compliance or quality failures. Input is a set of findings with file:line references. Must re-run verification after fixes and report what was changed.
Agent
Use this agent to implement a specific task from a plan. Follows the plan exactly, writes tests, self-reviews, commits, and reports back with evidence. Dispatched once per task — fresh agent each time to prevent context pollution.
Agent
Use this agent for Phase 1 investigation when the task requires external knowledge — API documentation, library usage patterns, current best practices, or technology comparisons. Returns a structured findings report with sources.
Agent
Use this agent for Phase 3 plan validation — independently verifies that a plan is correct, complete, and safe to implement. Reads the actual codebase to verify the planner's claims.
Agent
Use this agent for Phase 2 planning — takes investigation findings and produces an actionable change plan with exact file:line specifications. Every change must be specific enough for an implementor with no project context to execute.
Agent
Use this agent when you need to understand how an external library or open-source project implements a feature by examining actual source code. Clones the repo, investigates with codebase analysis, and reports with file:line citations.
Agent
Use this agent after implementation completes to verify the implementation matches its specification. Compares actual code changes against the planned changes — does NOT trust the implementor's report. Must be dispatched BEFORE code quality evaluators.
Agent
Run full QA suite on a Typst package before publishing. Use when preparing a package for Typst Universe submission.
Agent
Verify Typst document output against requirements. Use after compilation when you need to confirm content, structure, or visual layout correctness.