Verifies AI engineering output claims via adversarial sub-agents producing structured verdicts with confidence scores.
npx skills add https://github.com/samuelcastro/skills --skill super-truth
Please help me install this Agent Skill. Skill: super-truth Source: https://github.com/samuelcastro/skills/tree/main/super-truth Command: npx skills add https://github.com/samuelcastro/skills --skill super-truth
It prevents optimistic but unverified AI results from being treated as true by adding an adversarial check that asks whether the claimed outcome actually holds in reality.
Ask your agent to run the command: /super-truth verify <input> using the producer artifact you want grounded in reality (such as a commit SHA, review comment URL, test spec path, or visual-diff log).
High-intent search queries and answers about installing and using this skill.
You can verify AI code review findings by running an adversarial verification gate that spawns multiple sub-agents to reproduce or refute the claimed outcomes, returning a structured verdict with counterexamples and a confidence score.
Adversarial verification is a mechanism that attempts to disprove AI-produced engineering workflow outputs by checking whether claimed results correspond to real, user-visible outcomes rather than accepting optimistic assertions at face value.
You validate visual diff verdicts and QA spec results by passing the producer artifacts to parallel adversarial sub-agents, which attempt to identify concrete counterexamples and output a publish, halt, or escalate-to-human gate decision.
Yes, confidence gating evaluates feature-implementation completion reports by measuring sub-agent agreement and contradictions, producing a 0–100 confidence score that drives an automated publish, halt, or escalate-to-human decision.
This approach supports claims originating from code review findings, fix commits, QA spec artifacts, visual-diff verdicts, and feature-implementation reports, grounding them in reality by attempting to reproduce the underlying claims.
You should avoid using adversarial verification when you lack producer artifacts like commit SHAs or test spec paths, as the process requires spawning parallel sub-agents to actively attempt to refute specific claims with concrete counterexamples.
LucasSantana-DevGuide code review requests with Git SHAs and feedback actions.
jjacksonProcesses chestnut images and generates a PDF report of findings.
multicamCoordinate AI coding agents to execute work orders and integrate versioned code.
etylsarinValidate agent delegation outputs and produce PASS or FAIL verdicts.
