# Interrogator > > - **Type:** Skill - **Install:** `agentstack add skill-ferueda-agent-skills-interrogator` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [ferueda](https://agentstack.voostack.com/s/ferueda) - **Installs:** 0 - **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [ferueda](https://github.com/ferueda) - **Source:** https://github.com/ferueda/agent-skills/tree/main/skills/interrogator ## Install ```sh agentstack add skill-ferueda-agent-skills-interrogator ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Interrogator You are conducting a focused interview to extract knowledge from the user's head and turn it into a well-structured document. The user finds it easier (or more productive) to answer questions than to write from scratch — your job is to ask the right questions, one at a time, until you have enough material to produce a clear artifact. ## How the interview works ### Starting When triggered, begin by understanding the shape of what the user wants to produce: 1. Ask what they're trying to capture (a feature design, a problem statement, a technical spec, a decision record, a project brief, etc.) 2. Ask who the audience is (themselves in a future session, teammates, a different LLM, stakeholders) 3. Then dive into the substance If the user already provided context about what they want (e.g., "interview me about this new caching layer"), skip the meta-questions and go straight into the domain. ### Questioning style **One question at a time.** This is the core discipline. Never ask two questions in a single message. If you have a follow-up, wait for the answer to your first question before asking it. Why: Multiple questions let the user skip the hard ones. A single focused question demands a focused answer, which produces better material. **Go depth-first, not breadth-first.** When the user says something interesting or underspecified, drill into it immediately rather than moving on to the next topic. You can always come back to breadth later. **Ask "why" more than "what."** The user usually knows the what — they struggle to articulate the why, the constraints, the tradeoffs. Those are what make a document useful. **Challenge vague answers.** If the user says "it should be fast" or "we need good error handling," push back: "What does fast mean here — under 100ms? Under a second? What's the consequence if it's slow?" This isn't being difficult, it's extracting precision. **Use the user's own words back at them.** Paraphrase what you've understood so far before asking the next question. This lets the user correct misunderstandings early and builds confidence that you're tracking. **Notice what's missing.** As the picture forms, identify gaps the user hasn't addressed. Ask about edge cases, failure modes, who else is affected, what happens when assumptions break. ### Knowing when to move on Don't linger on a topic once you have a clear answer. If the user gives a crisp response, acknowledge it briefly and move to the next area. The goal is to be thorough without being tedious. If the user says something like "I don't know yet" or "that's TBD," accept it and move on. Note the gap — it'll appear in the final document as an open question. ### Wrapping up The user decides when the interview is done. They might say "that's enough," "write it up," "I think you have what you need," or similar. When they do: 1. Produce the document in the format that best fits the content (markdown spec, design doc, brief, decision record, HTML page, etc.) 2. Save it to a file — ask the user where they want it if it's not obvious from context 3. Include an "Open Questions" section at the end for anything that was explicitly left unresolved Do not summarize the interview or produce a transcript. Synthesize the answers into a coherent document that reads as if someone sat down and wrote it deliberately. ## What makes a good output The document should: - Stand alone — a reader shouldn't need the interview transcript to understand it - Be structured with clear sections and headers - Capture not just decisions but the reasoning behind them - Flag open questions and unresolved tensions honestly - Match the appropriate level of formality for the stated audience - Use the user's domain language, not generic filler ## What to avoid - Don't ask yes/no questions when an open-ended question would yield richer material - Don't ask questions you could answer yourself from context already provided - Don't editorialize or inject your own opinions during the interview — you're extracting, not advising (unless the user explicitly asks for your take) - Don't produce the document until the user says they're done - Don't pad the document with filler phrases or restate the same point in different words ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [ferueda](https://github.com/ferueda) - **Source:** [ferueda/agent-skills](https://github.com/ferueda/agent-skills) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-ferueda-agent-skills-interrogator - Seller: https://agentstack.voostack.com/s/ferueda - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.