G
GARLPROTOCOL
GARL FOR CODE · OPEN SOURCE

Cryptographic proof
for every AI-generated commit.

Nearly half of all new code on GitHub is AI-touched (Octoverse 2025). Who wrote it? Which model? Was it reviewed? GARL for Code signs every AI-authored commit with ECDSA-secp256k1 and makes provenance verifiable.

Install in 1 minute →SourceLive receipt example
pull/#1337
GARL Receipt· 3 of 5 commits signed
🔐 GARL Verified AI Code
├── Model: claude-opus-4-6
├── Tool: Claude Code
├── Files touched: 12
├── Duration: 4m 12s
├── Signed: ECDSA-secp256k1 ✓
└── Receipt: https://garl.ai/r/a8f3c2d1
↪ PR comment + informational check · posted by the action
The problem

AI is writing your code. Can you prove who wrote what?

Provenance gap

Claude Code, Cursor, Copilot, Aider — every modern IDE invites an agent into the commit. Git history captures the author but not the model, the prompt, or the verifier.

Compliance clock

CA · EU · ISO 42001

California SB 942 AI Transparency Act is live since 1 Jan 2026. The EU AI Act Code of Practice on AI-generated content lands in August 2026. ISO/IEC 42001 Annex B demands traceable AI logs. Audit-ready AI provenance is already a procurement requirement.

Reviewer fatigue

Reviewers need a concise signal that an AI assisted this diff, which tool, and when. A sticky PR comment with a verifiable receipt beats digging through commit trailers.

How it works

5 lines of YAML. Every AI commit gets a signed receipt.

.github/workflows/garl-receipt.yml
name: GARL Receipt
on:
  pull_request:
    types: [opened, synchronize, reopened]
jobs:
  sign:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write
      checks: write
    steps:
      - uses: actions/checkout@v4
        with: { fetch-depth: 0 }
      - uses: Garl-Protocol/[email protected]
        with:
          garl-api-key: ${{ secrets.GARL_API_KEY }}
          garl-agent-id: ${{ secrets.GARL_AGENT_ID }}

1. Detect

The action walks the PR's base..head commits and matches co-author trailers (Claude, Cursor, Copilot, Aider, Codex) + Claude Code markers + model-name heuristics. Confidence scored 0.4–1.0.

2. Sign

Every qualifying commit is submitted to GARL as a signed trace. ECDSA-secp256k1 signature + SHA-256 hash + immutable ledger record.

3. Report

One sticky PR comment summarizes `N of M commits signed · breakdown by tool`. A neutral (informational) `GARL Receipt` check run lists every receipt URL.

4. Audit

Each receipt URL (garl.ai/r/{short}) is a public proof card — reviewable by auditors, compliance officers, reviewers, or your future self.

New · GARL PR Bot

Don't want to touch YAML? Install the GitHub App.

One-click GitHub App posts the same sticky receipt comment on every PR — including fork PRs, which the Action workflow cannot authenticate. HMAC-gated, per-repo rate-limited, non-blocking. Perfect for org-wide rollouts without per-repo secret wrangling.

Add GARL PR BotApp manifest + self-host
What you get

A single workflow. Four compounding benefits.

Signed PR check

Informational `GARL Receipt` check run on every PR — never blocks, always visible.

Receipt URL

Public, shareable garl.ai/r/{short} page per commit, with agent and ECDSA proof.

Rich previews

Auto-generated 1200×630 Open Graph cards so receipts render beautifully in Slack, X, LinkedIn.

Audit trail

Immutable ledger record with who/what/when — exports for CA SB 942, EU AI Act Code of Practice, and ISO 42001 Annex B audits.

Install

1 minute, 2 secrets, 1 workflow file.

  1. 1. Register a repo agent — via the garl_register_agent MCP tool or curl:
    curl -sX POST https://api.garl.ai/api/v1/agents/auto-register \
  -H "Content-Type: application/json" \
  -d '{"name":"gh-<owner>-<repo>","framework":"github-action"}'
  2. 2. Save the returned agent_id and api_key as GitHub secrets GARL_AGENT_ID and GARL_API_KEY.
  3. 3. Drop the workflow above into .github/workflows/garl-receipt.yml. Open a PR that contains an AI co-author trailer — the action signs it.
Open source · Apache 2.0

Built on GARL Protocol — the open trust layer for AI systems. Source on GitHub. Python & JS SDKs, MCP server (12 named tools + batch variants), and the full REST API are all part of the same monorepo. No SaaS lock-in, no black-box scoring — everything verifiable.