Triage

Analyze requirements, issues, or specifications for completeness. Explore the codebase to understand context, then generate targeted questions to fill gaps before implementation begins.

Workflow

1. Receive and Parse Input

The user provides input as plain text - typically:

• Issue descriptions
• Acceptance criteria
• Feature requests
• Bug reports with reproduction steps
• Stakeholder comments or requirements

Extract and identify:

• Goal: What is being requested or solved
• Scope: What parts of the system are affected
• Constraints: Any mentioned limitations or requirements
• Acceptance criteria: How success will be measured

2. Explore Codebase and Documentation

Codebase

Use the Explore agent to understand:

• Project structure and architecture
• Relevant existing code that would be modified
• Patterns and conventions used in the codebase
• Related features or similar implementations
• Test coverage and testing patterns

Local Documentation

Search for relevant documentation that may contain implicit requirements:

• docs/, documentation/, wiki/ directories
• README.md, CONTRIBUTING.md, ARCHITECTURE.md
• API specs (openapi.yaml, swagger.json)
• ADRs (Architecture Decision Records)
• Inline documentation and code comments

External Documentation

If configured (see config.md), search external documentation systems for:

• Business rules that the ticket assumes but doesn't state
• Existing constraints documented elsewhere
• Related features that may have dependencies
• Domain terminology that clarifies ambiguous terms
• Edge cases already documented for similar features

3. Assess Completeness

Evaluate the input against these dimensions:

Dimension	Questions to Consider
Problem clarity	Is the problem/goal clearly stated? Why is this needed?
Scope definition	What's in scope? What's explicitly out of scope?
User impact	Who benefits? What user journey is affected?
Acceptance criteria	How do we know when it's done? What are success metrics?
Edge cases	What happens in error states? Empty states? Boundaries?
Technical scope	Which components/files are affected? API changes?
Dependencies	Blocked by anything? Needs coordination with other work?
Non-functional	Performance requirements? Security considerations?
Data	Schema changes? Migration needed? Data implications?
UX/UI	Designs provided? Interaction patterns defined?

Rate overall completeness: Ready / Mostly Ready / Needs Clarification / Underspecified

4. Generate Questions

Identify questions organized by:

1. Blockers - Questions that must be answered before any work can begin
2. Scope clarification - Questions that define boundaries
3. Technical decisions - Questions about implementation approach
4. Nice to know - Questions that would help but aren't blocking

5. Ask Questions Interactively

Use AskUserQuestion tool to ask the user any questions they might be able to answer (especially blockers and scope questions).

• Ask 1-4 questions at a time using the tool's multi-question format
• For each question, provide 2-4 concrete answer options
• Record answers to incorporate into the output
• If user selects "Other" without an answer or says they don't know → mark as unanswered

Only questions the user couldn't answer go to the output file for stakeholder follow-up.

6. Write Output

Output location: If the project defines an .aiwork/ folder protocol (e.g., naming conventions, frontmatter, folder structure), follow that protocol. Otherwise use these defaults:

mkdir -p ./.aiwork/{date}_{slug}

Save to ./.aiwork/{date}_{slug}/triage.md

Where:

• {date} = current date as YYYY-MM-DD
• {slug} = with ticket ID: <ticket-id>-<slugified-title>, without: <slugified-title>

Slugify: lowercase, spaces→hyphens, remove special chars, max 40 chars.

If a task folder already exists for this ticket/slug (search .aiwork/ for matching folders), place the file there instead of creating a new one.

Output Format

# Triage: [Short title]

**Source**: [Ticket ID/URL if available, or "Manual input"]
**Date**: [YYYY-MM-DD]
**Completeness**: [Ready | Mostly Ready | Needs Clarification | Underspecified]

## Summary

[1-sentence summary of what was provided]

### Understanding

[2-3 sentences capturing the core request and affected areas based on codebase exploration]

### What's Clear

- [Bullet points of well-defined aspects]

### Implicit Requirements (from docs)

- [Requirements found in documentation that the ticket assumes but doesn't state]
- [Business rules, constraints, or edge cases documented elsewhere]

### Gaps Identified

- [Bullet points of missing or ambiguous information]

### Questions

#### Blockers

1. [Question] — [Why this blocks progress]

#### Scope Clarification

1. [Question]

#### Technical Decisions

1. [Question]

#### Nice to Know

1. [Question]

Note: Only include questions the user couldn't answer during interactive clarification.

Tips

• Don't ask questions the codebase or docs already answer - explore first
• Surface implicit requirements from docs - tickets often assume documented knowledge
• Prioritize ruthlessly - 3 critical questions > 10 nice-to-haves
• Frame questions to unblock decisions, not gather trivia
• When docs contradict the ticket, flag it as a blocker question