interrogate

Files (6)

• SKILL.md
• openai.yaml · agents/
• code-quality-review.md · references/
• lead-judgment.md · references/
• reviewer-prompt.md · references/
• rubric.md · references/

SKILL.md (first lines)

# Interrogate

Adversaries try to break the requested change or design. A finding survives only when a run proves it. Report findings; do not apply fixes.

1. Pin scope: what the user names, otherwise the branch diff from its merge base through the working tree. State the intent per [in-your-own-words](../in-your-own-words/SKILL.md) from the request, commits, and PR description. Ask when the intent is unclear.
2. Give each of 2-3 adversaries its own detached worktree outside the target checkout, with the scope's uncommitted changes copied in, untracked files included. Build patches with `git diff --no-ext-diff --binary`, since an external diff tool yields no applicable patch. Run them through the [selected execution method](../z-mode/references/native-hosts.md#execution-selection) and record actual identities; same-model independence is not provider diversity. When the [other-provider route](../z-mode/references/native-hosts.md#other-provider-route) is available and offers a model family other than yours, run one adversary through it in that adversary's worktree. The target checkout and its unrelated work stay untouched.
3. Brief each adversary with the scope, the intent, and: "Break this change. Prove each finding with a run against the real code: a failing test, a repro script, or command output showing the wrong behavior. For a design or plan, run a prototype or probe the current tree. Keep repros in your own worktree. Report each finding with a title, the exact command, the output it printed, and the output the intent requires. Leave out anything you did not run, including style, naming, structure, and risks that need a future change to matter. No findings is a valid result."
4. Wait for terminal results. Rerun each finding's command yourself in a fresh isolated worktree at the scope, carrying over only the repro files it names. Keep a finding when the output shows the failure and the failure breaks the intent; drop the rest. Merge findings that share a root cause. Keep every verified finding, not one adversary's list.
5. Remove the worktrees you created.

Return:
- **Intent**: the restatement.
- **Findings**: for each verified finding, the command, its output, how it breaks the intent, and the adversaries that found it. Say "no verified findings" when none survive.
- **Dropped**: for each adversary, the count of dropped findings and why.

If agents are unavailable, label a parent-only interrogation. A finding nobody could run stays dropped, not promoted on argument.

From fullerzz/agent-skills, MIT.

More from fullerzz/agent-skills