Skip to main content
Task specifications are optional Markdown briefs for work that benefits from stable goals, constraints, and acceptance criteria. They do not require a parser or a specific AI runtime. Store reusable templates under .ai/tasks/templates. You can copy a template into an issue, a pull request, an AI conversation, or a temporary task file.

Choose a template

Each template links to a matching workflow under .ai/workflows.

Required context

A useful specification contains:
  • Goal: the observable outcome, not an implementation guess.
  • Context: owning area, current behavior, related paths, and affected platforms.
  • Constraints: interfaces, boundaries, compatibility, and behavior that must remain stable.
  • Acceptance criteria: independently verifiable outcomes.
  • Validation: targeted commands and manual checks.
Make repository paths explicit. An instruction such as “update the profile flow” is ambiguous; a path such as apps/mobile/app/(app)/(tabs)/profile.tsx gives the AI a concrete starting point.

Example

Use a task with an AI tool

Ask your AI tool to read the task, AGENTS.md, and .ai/README.md, then follow the linked workflow. Review its plan before allowing implementation for broad or ambiguous work.
Keep credentials, customer data, private endpoints, and machine-specific paths out of committed task files.