# Ac Interrogation

> Interrogate an idea, plan, or decision before commitment: structured questioning that surfaces assumptions, maps failure modes, tests evidence, forces alternatives, and closes without looping. Eight leaves: interrogation-protocol (session order, cold read, time-box), question-craft (past behaviour over hypotheticals, follow-ups), assumption-audit (load-bearing assumptions, confidence labels), failure-mapping (pre-mortem, inversion, base rates), evidence-dial (falsification test, decide-vs-research gate), alternative-forcing (three alternatives plus do-nothing), steelman-discipline (attack the strongest version), close-and-stop (stop criteria, anti-loop cap, assumption ledger, kill criteria). Use when asking 'grill me on my plan', 'stress-test this idea', 'poke holes in this', 'interrogate this decision', 'what am I missing', 'challenge my assumptions'. Not for: reviewing an existing artifact (use ac-code-review), open-ended research (use ac-deep-researcher), or writing requirements (use ac-spec-writing).

- Skill: `jiayaoqijia/ac-interrogation` (Agent Skill, multi-file: 12 files)
- Install (CLI): `npx get-skill add jiayaoqijia/ac-interrogation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/jiayaoqijia/ac-interrogation/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- License: MIT
- Author: jiayaoqijia (https://skillmd.com/u/jiayaoqijia)
- Updated: 2026-11-16
- Page: https://skillmd.com/skills/jiayaoqijia/ac-interrogation

---


# Interrogation

## Overview

The craft of interrogating an idea, plan, spec, or decision before commitment: the "grill me" pattern. The reader runs a structured session that reads the plan cold, surfaces assumptions, tests evidence, maps failure modes, forces alternatives, and then closes on purpose. The output is a revised plan plus a written record, not a verdict from the questioner.

## When to use

- Someone asks to be grilled: "grill me on my plan", "stress-test this idea", "poke holes in this", "what am I missing".
- A decision is approaching commitment and nobody has attacked it yet.
- An idea feels settled too early, or its author cannot state what would change their mind.

## When NOT to use

Reviewing an existing artifact is outside; ac-code-review covers that. Open-ended research is outside; ac-deep-researcher covers that. Writing requirements is outside; ac-spec-writing covers that. Routine two-way-door choices with small stakes need one question, not a session.

## Subskill index

| Leaf | domain segment | scope (one line) |
|---|---|---|
| `subskills/interrogation-protocol.md` | `critical-thinking/interrogation-protocol` | Session order and pacing: purpose, assumptions, evidence, failure modes, alternatives, commitment; serial vs batch; time-box |
| `subskills/question-craft.md` | `critical-thinking/question-craft` | How to ask: past behaviour over hypotheticals, "what would have to be true", open vs leading questions, follow-up probes, silence |
| `subskills/assumption-audit.md` | `critical-thinking/assumption-audit` | Inventory unstated assumptions; separate load-bearing from incidental; premise check; cheap tests; confidence labels |
| `subskills/failure-mapping.md` | `critical-thinking/failure-mapping` | Pre-mortem, inversion, weakest link, reference-class base rates, and the failure that actually ends the plan |
| `subskills/evidence-dial.md` | `critical-thinking/evidence-dial` | Evidence that exists vs evidence needed; falsification test; decide-vs-research-more gate; calibrated confidence |
| `subskills/alternative-forcing.md` | `critical-thinking/alternative-forcing` | At least three alternatives including do-nothing; one-way vs two-way doors; opportunity cost; fake alternatives |
| `subskills/steelman-discipline.md` | `critical-thinking/steelman-discipline` | Steelman before attacking; attack the strongest version; separate person from proposition; straw-man and motte-and-bailey traps |
| `subskills/close-and-stop.md` | `critical-thinking/close-and-stop` | When to stop: repeated answers, diminishing returns, decision ready, fatigue; anti-loop cap; assumption ledger; kill criteria |

## Status and honest limits

- Seedling: built from the public canon cited in each leaf; the full session order has not yet been stress-tested across many live grilling sessions.
- The protocol produces better questions and a written record. It does not produce certainty; the plan owner still decides.
- Questioning works on a willing subject. Against someone defending a fixed position, the session can only surface the disagreement, not resolve it.

