Repository navigation
Conversation
Adversarial review: adversaries try to break a change or design and only findings proven by a run against the real code survive. Adapted from the pstack skills (MIT, (c) Lauren Tan, https://github.com/cursor/plugins); the subagent fan-out was rewritten to be harness-agnostic.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Part of a 10-PR series contributing the agent-guardrail skill stack (prove-it-works, blast-radius, fix-root-causes, subtract-before-you-add, correct, interrogate, show-me-your-work, create-verification-skill, maintain-verification-skill, property-based-testing). Each PR is independent, adds exactly one single-skill plugin, and can be reviewed and merged in any order; adjacent additions to
marketplace.json/README.mdmay show trivial conflicts with sibling PRs.What
Adds the
interrogateplugin: adversarial review where adversaries try to break a change or a design, and a finding survives only when a run proves it. Fans out several adversary subagents (ideally on different model families), merges findings, and reruns every finding's command before reporting. Findings are reported, never applied.Why
The marketplace ships code-review, security, and runtime-verification guardrails (
droid-control's/verify), but nothing covering adversarial verification discipline for the agent itself. This stack was built and battle-tested in a hackathon repo's quality gates; we are contributing it back.Provenance and license
Adapted from the pstack skills by Lauren Tan (cursor/plugins, MIT License, commit
c742dd6). The subagent fan-out was rewritten to be harness-agnostic (no Cursor arena/model-role machinery); the prove-findings-by-run contract is unchanged. Made model-invocable so Droid applies it proactively.Checklist
plugins/interrogate/with.factory-plugin/plugin.jsonskills/(per the check-skills CI).factory-plugin/marketplace.jsonentry added