When software is produced by autonomous agents, traditional git blame ('git blame file.ts') loses its utility. Knowing that an agent modified line 42 does not reveal which model was used, what prompt guided the modification, or what security checks were bypassed. SUTRA's Cryptographic Commit Provenance solves this transparency crisis.
The Anatomy of an Attested Commit#
In SUTRA, every commit produced by an agent is coupled with a canonical cryptographic attestation packet. This packet binds the git tree hash with:
- Agent Registration ID: The unique cryptographic fingerprint of the authorized agent.
- Human Session Authority: The human engineer who initiated or authorized the agent task.
- Prompt Fingerprint: A SHA-256 hash of the exact prompt, system instructions, and file context.
- Automated Verification Hash: Test suite outcomes, linter results, and security scanner outputs.
Invalidating Approvals on Synchronize Events#
A common vulnerability in multi-agent workflows is the 'bait-and-switch' push: an agent receives approval on a benign commit, and then pushes a secondary commit containing malicious or untested changes. SUTRA binds approvals strictly to the commit HEAD SHA. Any new push automatically invalidates all previous approvals.
Frequently Asked Architectural Questions
#Commit Provenance#Cryptography#Git#Auditability#Security