ai-truthfulness-enforcer_skill

Overview

This skill enforces mandatory verification before any success or progress claim is allowed. It intercepts claim language, requires live testing and cryptographic evidence, and rejects statements that lack provable proof. The goal is to eliminate fabricated results and ensure every claim is backed by reproducible artifacts.

How this skill works

The skill scans outgoing messages for trigger phrases and high-suspicion patterns, halting any message that asserts success or progress. It then runs mandatory evidence collection workflows (build runs, browser tests, logs, screenshots) and produces SHA256 evidence hashes and timestamped chain-of-custody records. If evidence is missing, inconsistent, or tampered with, the skill automatically rejects the claim and initiates an audit and lockdown sequence.

When to use it

• Any time a response includes success, completion, or progress language (e.g., "fixed", "done", "X% complete").
• Before reporting build, test, or deployment outcomes to users or logs.
• When asserting functionality, performance improvements, or error count reductions.
• Whenever a claim could affect user decisions, releases, or public reports.
• During automated pipelines that publish status or metrics to stakeholders.

Best practices

• Run live, reproducible commands for each claim (e.g., npm run build, npx vue-tsc).
• Capture full console output with timestamps and attach terminal screenshots or video.
• Use automated browser tests (Playwright) for functionality claims and include before/after screenshots.
• Compute and store a SHA256 hash of claim|evidence|timestamp and verify chain-of-custody.
• Ask the user to verify human-observed fixes before marking any issue as complete.

Example use cases

• CI pipeline that blocks merge notes claiming "build passed" unless live build and logs are attached.
• Developer assistant that must produce verified screenshots and console logs before declaring a UI bug fixed.
• Release notes generator that only publishes performance gains after repeated benchmark runs and independent metric verification.
• Support workflow that requires end-user confirmation via a specific AskUserQuestion prompt before closing tickets.
• Audit tool that records tamper-evident evidence for all claims made by an automated agent.