Guides a one-question-at-a-time design interview, captures alignment in agent/EVE-BRIEF.md, then scaffolds and implements a runnable eve agent with verbose teaching comments.
SkillScore74Maintained
Automatic
◆AI Engineer pack
Last verified 8/20/2026
What it changes
Passed 13 of 18 test cases with the skill, and 9 of 18 without it.
Without skill48%
With skill74%
Checks it passed that the model alone missed
The response specifies openai/gpt-4.1-mini as the default model string.
The response specifies running npx eve@latest init . in the project root.
The response requires AI_GATEWAY_API_KEY to be listed inside .env.example.
and 7 more in the full report
9 fixed8 passed either way3 still failing3 broke
Pass-rate lift
+26 pp
Tokens
+88%
Agent turns
0%
Response time
-62%
Model gemini-3.6-flashMeasured 8/20/2026v117 of 23 cases passed with skill · 18 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 9 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 9
Prompt
When preparing integration secrets for our new agent, what example environment file must be created, and what default gateway key must it contain?
What was checked, without and with the skill
Checked
Without skill
With skill
The response specifies creating a .env.example file.
Passed
Passed
The response requires AI_GATEWAY_API_KEY to be listed inside .env.example.
Missed
Passed
case-11 · gemini-3.6-flash · 8/20/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md185 lines2 files
create-agent
Turn a rough agent idea into a runnable eve agent through structured Q&A, a written brief, then scaffold + implementation.
Phases
Task Progress:
- [ ] Phase 1 — Design interview (one question at a time)
- [ ] Phase 2 — Write agent/EVE-BRIEF.md and get confirmation
- [ ] Phase 3 — Scaffold eve, implement, verify
Do not skip Phase 1 or Phase 2. Do not scaffold or write agent code until the user confirms the brief.
Phase 1 — Design interview
Interview the user until you have a shared, implementable picture of the agent.
Rules
One question at a time. Wait for the user's answer before the next question.
Include a recommended answer with every question — state your pick and why in one or two sentences.
Explore instead of asking when the answer is in the repo, docs, or env (existing agents, package.json, API keys, monorepo layout).
Follow the decision tree. Resolve blockers before dependent questions (purpose → surfaces → tools → scope → runtime).
Use concrete scenarios. When behavior is fuzzy, invent a user message and ask what the agent should do step by step.
Tighten vague language. Replace overloaded words ("handle", "manage", "process") with specific capabilities or tool names.
Surface contradictions. If a new answer conflicts with an earlier one, call it out before moving on.
What to cover (in rough order)
Area
Goal
Purpose
One sentence: what job does this agent do?
Users & entry
Who talks to it, and through what channel(s)?
Core loop
Typical turn: user says X → agent does Y → user sees Z