| name | create-eve-agent |
| description | Use when adding a new Eve app, local Eve subagent, or reusable Eve agent package in this repository. |
Create Eve Agent
This is a design-first skill. Do not start implementation in the same session unless the user
explicitly overrides that after reviewing the plan.
Start with alignment. Ask the structured questions in references/questions.md before proposing
files unless the answers are already written in the task.
Then choose the smallest Eve surface that matches the requirement:
- Skill: optional guidance.
- Tool: deterministic typed execution.
- Hook: lifecycle context seeding/checking or stream-event observation.
- Subagent: separate specialist prompt, tools, and context.
- Package: behavior shared by multiple apps.
Write the plan/checklist through the paths configured in .ai/config.json. If an imported template
or instruction mentions another folder, map it through .ai/config.json first.
Use templates/app-checklist.md for a new app and templates/subagent-checklist.md for a local
subagent. Keep fixtures synthetic and app-local unless shared context is explicitly allowed. Default
subagents to app-local unless immediate reuse clearly justifies a shared package.
Write an ADR only when the design creates a durable architectural decision, such as a new app
boundary, shared package, subagent extraction, auth/session ownership model, or non-obvious
deterministic boundary.
End the session with explicit instructions for how to trigger implementation, for example:
To implement this plan, start a new session and say:
"Implement .ai/plans/<plan-file>.md. Do not run pnpm eval unless I explicitly ask."
Mention the affected verification commands that implementation should run. Keep pnpm eval clearly
marked as opt-in/model-backed.