Interview Dev Loop is an Agent Skill for the workflow we call an "interview implementation loop".
The agent investigates first, asks one multiple-choice clarification question at a time, writes a plan only after material ambiguity is resolved, waits for approval, then runs implementation and review until no P0/P1/P2 findings remain.
The internal Japanese name is インタビュー実装ループ.
This repository is a public, agent-agnostic version of the original interview-plan-implement-review-loop skill. Product-specific repo rules have been generalized, but the core workflow is the same.
- Investigates code, docs, tests, existing plans, and local instructions before planning.
- Surfaces findings in a fixed shape before asking questions.
- Asks one A/B/C clarification question at a time.
- Writes a durable plan only after material ambiguity is resolved.
- Waits for user approval before implementation.
- Runs development and review as a loop.
- Continues until no unresolved P0/P1/P2 findings remain.
Use this for feature work, bug fixes, and product tasks where the request is real but underspecified.
Good fit:
- "Add this integration."
- "Implement this ticket."
- "Build this feature, but ask before making product-scope decisions."
- "Plan first, then implement after approval."
Poor fit:
- Tiny edits where the next action is obvious.
- Pure explanation or research tasks.
- PR-only review loops after a pull request already exists.
- Tasks where the human has already supplied a complete implementation plan.
Codex can use Agent Skills directly, but reusable distribution is best packaged as a plugin.
Add this repository as a plugin marketplace:
codex plugin marketplace add unsu0707/interview-dev-loopThen install interview-dev-loop from the Codex plugin directory.
For other coding agents, ask the agent you use:
Install the [unsu0707/interview-dev-loop](https://github.com/unsu0707/interview-dev-loop) skill.
Explicitly invoke the skill when starting a coding task:
Use interview-dev-loop for this task: add support for ...
Or mention it by name in tools that support skill mentions:
$interview-dev-loop Add support for ...
The first useful response should be an investigation summary and, if needed, one clarification question. It should not jump directly into implementation.
.
├── .codex-plugin/plugin.json
├── .agents/plugins/marketplace.json
├── marketplace.json
├── skills/interview-dev-loop/SKILL.md
├── skills/interview-dev-loop/agents/openai.yaml
├── skills/interview-dev-loop/references/
└── examples/
This repository follows the open Agent Skills format:
- one directory per skill
SKILL.mdwithnameanddescription- lowercase hyphenated skill name
- optional
references/for progressive disclosure
It is intended for Codex, OpenCode, Claude Code, and other Agent Skills-compatible coding agents.
MIT