pot-ci — ThoughtProof Code Verification for GitHub Actions

Multi-model epistemic verification for AI-generated code. Stop merging code that only one model reviewed.

What it does

pot-ci runs every PR through ThoughtProof's epistemic verification pipeline:

1. 3+ AI models independently review your code diff
2. An adversarial critic tears their reviews apart
3. A synthesizer produces a final assessment with confidence score
4. Results posted as a PR comment with pass/fail status

If confidence drops below your threshold, the check fails — blocking the merge.

Quick Start

# .github/workflows/verify.yml
name: AI Code Verification
on: [pull_request]

jobs:
  verify:
    runs-on: ubuntu-latest
    permissions:
      pull-requests: write
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - uses: thoughtproof/pot-ci@v1
        with:
          threshold: '0.7'
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          XAI_API_KEY: ${{ secrets.XAI_API_KEY }}
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
          DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}

PR Comment Example

✅ ThoughtProof Verification: PASSED

Metric	Value
Confidence	84.2%	████████░░
MDI (Model Diversity)	72.0%	███████░░░
Threshold	70.0%
Duration	12.3s

Model Perspectives

Model	Position	Confidence
xAI (grok-4-1-fast)	🟢 support	89%
DeepSeek (deepseek-chat)	🟡 partial	72%
Moonshot (kimi-k2-turbo)	🟢 support	91%

Inputs

Input	Default	Description
threshold	0.7	Minimum confidence to pass (0.0-1.0)
fail-on-low-confidence	true	Fail the check if below threshold
comment	true	Post results as PR comment
generators	grok,deepseek,kimi	Comma-separated generator models
critic	claude-sonnet	Adversarial critic model
synthesizer	deepseek	Synthesis model
max-diff-lines	500	Truncate larger diffs

Outputs

Output	Description
confidence	Confidence score (0.0-1.0)
mdi	Model Diversity Index (0.0-1.0)
result	pass, fail, or skip
block-hash	SHA-256 hash of the epistemic block

Available Models

Alias	Provider	Env Variable
grok	xAI	XAI_API_KEY
deepseek	DeepSeek	DEEPSEEK_API_KEY
kimi	Moonshot	MOONSHOT_API_KEY
claude-sonnet	Anthropic	ANTHROPIC_API_KEY
claude-opus	Anthropic	ANTHROPIC_API_KEY

BYOK — Bring Your Own Keys

pot-ci never sees your code or API keys. Everything runs in your GitHub Actions runner:

• Code stays in your CI environment
• API keys stored in GitHub Secrets
• No ThoughtProof servers involved
• Models are called directly from the runner

How it works

PR opened → pot-ci triggers
    ↓
Git diff extracted (truncated to max-diff-lines)
    ↓
AI markers detected (Copilot, Cursor, Claude, etc.)
    ↓
3 generators independently review the diff
    ↓
Adversarial critic evaluates all perspectives
    ↓
Synthesizer produces final assessment
    ↓
PR comment posted + check pass/fail

Self-hosted models

Use Ollama or any OpenAI-compatible endpoint by forking this action and modifying src/models.ts. Full local verification is possible — see thoughtproof.ai/blog/eu-ai-act-compliance for data residency details.

License

MIT — ThoughtProof | Patent Pending