/pstack:interrogate
AutoIn one lineHas several different models hunt for bugs in your change, keeping only ones a command can show. It changes nothing.
- When to use
- The code is written but you're not confident, and want it picked apart before merging.
/pstack:interrogate Review the whole branch and be picky. Only report real bugs or things that used to work and now don't.Just asking vs. using it
The same request, typed plainly or with /pstack:interrogate. Here's what changes.
With /pstack:interrogateWhat the document says
4 sections, about 350 words, 2 min to read. Click a section to read it.
OpeningIt says adversaries try to break the change, only findings proven by a run count, and it reports without fixing.
- 1FrameDefaults to the whole branch diff, restates the intent with
in-your-own-words, and asks you when the intent is unclear. - 2Fan outReuses
arena's spawning steps, giving each adversary its own worktree and the same "break this change" brief. - 3VerifyReruns each finding's command, keeps it only if it fails and breaks the intent, and merges shared root causes.
- 4OutputReports three parts: the restated intent, each verified finding with who found it, and each adversary's dropped count and reasons.
Wording worth borrowing
A few sentences from the original: what each one makes the AI do, why it works, and how you can borrow it.
Cast the AI as an opponent
Adversaries try to break a change or a design.
- Makes the AI
- The helpers aren't there to vouch for the code. Their job is to break it.
- Why it works
- "Adversary" and "break" turn "take a look" into "defeat it", so the AI hunts harder.
- Borrow it
- When the AI reviews a plan, say: "You're the toughest competitor. Prove this plan won't work."
Put actual next to expected
Report each finding with a title, the exact command, the output it printed, and the output the intent requires.
- Makes the AI
- Each problem comes with the command, what it printed, and what it should print, so you can rerun it.
- Why it works
- Writing "what happened" next to "what should happen" shows the gap at once and leaves no room for hand-waving.
- Borrow it
- When the AI reports bugs, write: "For each: what you did, what you saw, what you expected to see."
Name what the rule covers
Leave out anything you did not run. That covers style, naming, structure, and risks that need a future change to matter.
- Makes the AI
- It won't pad the list with naming, style, or "might break later". Every finding was actually run.
- Why it works
- "Leave out what you didn't run" leaves gaps. Naming the kinds most likely to sneak in closes them.
- Borrow it
- When the AI hunts for problems, write: "Only report what stops a customer paying. Colors, wording, and future risks don't count."
Say "nothing" is a fine answer
No findings is a valid result.
- Makes the AI
- When the code is fine, it says so instead of inventing problems to look busy.
- Why it works
- An AI asked to find problems feels it must find some. Saying an empty result counts removes that pressure.
- Borrow it
- When the AI checks your work, add: "If you find no typos, just reply 'none found'."
Full text
The SKILL.md the AI reads, as written.
descriptionUse for "interrogate", "adversarial review", "multi-model review", "challenge this", "stress test this code", "find blind spots", or "tear this apart". Adversaries try to break a change or design, and only findings proven by a run survive.
Interrogate
Adversaries try to break a change or a design. A finding survives only when a run proves it. Report the findings. Do not apply fixes.
Frame
The scope is what the user names. Otherwise it is the branch diff from its merge base with the base branch, through the working tree. State the intent with the in-your-own-words skill, from the user's message, the commits, and the PR description. Ask the user when the intent is unclear.
Fan out
Run Phase A steps 3 and 4 and Phase B of the arena skill, with four changes. Read the interrogate reviewers line in place of arena runners. Copy the scope's uncommitted changes into each worktree. Spawn each adversary as a general-purpose subagent (generalPurpose in Cursor). In place of the arena task and rationale, give each adversary the scope, the intent, and this brief:
Break this change. Prove each finding with a run against the real code, such as a failing test, a repro script, or command output that shows 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. That covers style, naming, structure, and risks that need a future change to matter. No findings is a valid result.
Verify
Verify each finding per arena Phase F by rerunning its command. 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.
Skip arena Phases C, D, and E. Adversaries find different failures, so the verdict keeps every verified finding, not one adversary's list.
Output
- 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.
Original by Lauren Tan (poteto), MIT license. Checked against the pstack mirror at 97de40d. View on GitHub