Requirements
Detailed, actionable requirements derived from your task description and grounded in your project's existing patterns and constraints.
AI Workloads
Task Refinement uses AI to enrich raw task titles with structured requirements, acceptance criteria, edge cases, and implementation hints — automatically, before your team writes a single line of code.
Default output
From a single task title, the AI produces a complete, structured specification — ready for your team to act on.
Detailed, actionable requirements derived from your task description and grounded in your project's existing patterns and constraints.
A testable checklist that defines done — unambiguously. No more half-shipped features because the definition of complete was unclear.
Boundary conditions and failure modes surfaced before any code is written — saving hours of debugging and incident response later.
Unit, integration, and end-to-end testing scenarios tailored to the scope of your task — so QA and developers align before implementation begins.
Concrete pointers to relevant modules, patterns, and APIs in your codebase — informed by your source code and user input, so developers know exactly where to start.
Before & after
See what Task Refinement produces from a raw task title with no additional context provided.
Add user export functionality
## Summary
Implement a user data export endpoint that lets users download
their account data in CSV or JSON format, compliant with
GDPR Article 20.
## Requirements
- Endpoint accepts a `format` query parameter (`csv` | `json`)
- Includes user profile, activity log, and account settings
- Rate-limited to 1 request per hour per user
- Files >10 MB are streamed rather than buffered in memory
## Acceptance Criteria
- [ ] GET /api/users/{id}/export?format=csv returns valid UTF-8 CSV
- [ ] GET /api/users/{id}/export?format=json returns valid JSON
- [ ] Exceeding rate limit returns 429 with Retry-After header
- [ ] Export covers all personal data per GDPR Article 20
## Edge Cases
- User with zero activity records
- Concurrent export requests from the same account
- Export triggered during an ongoing account deletion
## Test Plan
- Unit: CSV and JSON serializers for all field types
- Integration: rate-limiter middleware under concurrent load
- E2E: complete export → download flow in staging
## Implementation Hints
- Extend `UserRepository.findByIdWithRelations()` in
`src/repositories/user.ts` to include the activity log
- Reuse `StreamResponse` from `src/http/responses.ts`
for large file delivery
- Rate-limiting middleware is in `src/middleware/rateLimit.ts`The refined spec is generated in seconds and fully editable before your team acts on it.
Customization
The defaults work out of the box. When you need more control, every part of the refinement process is customizable.
Define the exact structure of a refined task using your own Markdown template. Match your team's issue format precisely — Jira-style, Linear-style, or your own.
Override the default refinement prompt with your own instructions. Tell the AI to focus on security considerations, performance constraints, or domain-specific terminology.
Set different templates and instructions per project or override them for an individual task. Flexibility without extra friction.
Task Refinement is always opt-in. Skip it for a single task or disable it project-wide — you stay in control of what the AI touches.
Join the early access program and let Orchestrator's Task Refinement do the spec work so your team can focus on building.