spec-driven

Collegium spec reviewer — finds what will break. Identifies failure scenarios, edge cases, lifecycle state leaks, and untested assumptions before implementation.

Use this agent when you need to review recently written or modified code for quality, structure, and best practices. Invoke after completing a logical chunk of implementation, before committing, or when explicitly asked to review. Also use when reviewing a plan's proposed file structure before implementation.

Pure diagnostic agent for deep root-cause analysis. Use when investigating any bug, error, failure, or unexpected behavior — frontend, backend, infrastructure, or cross-cutting. This agent NEVER writes or modifies code. It only investigates and returns a diagnosis. Pass it: (1) the problem / symptoms, (2) reproduction steps if known, (3) relevant file paths, error messages, or logs. Use proactively whenever a user reports something broken or behaving unexpectedly.

Use this agent when you need comprehensive research on a specific topic, technology, or problem domain. This agent excels at gathering information from multiple sources, synthesizing complex information, and preparing detailed research outputs that other AI agents can use as foundation for their work. <example> Context: The user needs to understand a new technology before implementing it. user: "I need to implement WebRTC for our video chat feature. Can you research the key concepts and implementation considerations?" assistant: "I'll use the deep-research-analyst agent to gather comprehensive information about WebRTC." </example> <example> Context: The user is evaluating different architectural patterns. user: "We're deciding between microservices and monolithic architecture for our new project. I need a detailed comparison." assistant: "Let me engage the deep-research-analyst agent to research both architectural patterns thoroughly." </example> <example> Context: The user encounters an unfamiliar error or issue. user: "We're getting CORS errors in production but not in development. I need to understand why this happens and all possible solutions." assistant: "I'll use the deep-research-analyst agent to investigate CORS issues and their solutions comprehensively." </example>

Collegium spec reviewer — evaluates how the design fits with the existing system. Catches boundary violations, data flow issues, blast radius problems, and integration mismatches.

Deep code quality review focused on cohesion, understandability, editability, testability, and debuggability. Produces an impact/effort matrix with actionable findings.

Run pre-spec reconnaissance (prep), load an existing spec to resume work, update progress after implementation, review with the collegium panel, or create a new one. Use when starting a session around a feature, when the user mentions a spec by name, when asked to prep, scope, or review/critique a spec, or after completing implementation work.

Universal structural quality principles for code and plans. Evaluates mechanism vs business logic separation, file/function size, composability, and reusability. Works before writing code (on plans) or after (on implementations). Trigger on "review structure", "check structure", "structural review", or when code/plans need decomposition review.