Hi — 12-factor-agents is a great framework for thinking rigorously about what makes agents production-ready. I wanted to raise a pattern that might be worth adding or referencing: tamper-evident decision audit.
The 12 factors cover a lot of great ground — statelessness, human-in-the-loop, tool boundaries, etc. One pattern that seems missing is: how do you prove to an auditor what an agent actually decided and executed, after the session ends, without relying on mutable log files?
We built HELM AI Kernel to address exactly this. For every ALLOW/DENY/ESCALATE decision, HELM produces a cryptographically signed receipt and bundles them into an offline-verifiable EvidencePack — a tamper-evident bundle of what ran, when, and under which policy.
This matters for:
- Regulated industries (SOC2, HIPAA, finance) where you need to prove policy was enforced
- Incident forensics where logs may have been modified
- Auditor reviews of autonomous agent sessions
Quick test (~5 min):
brew install mindburnlabs/tap/helm-ai-kernel
helm-ai-kernel serve --policy ./release.high_risk.v3.toml
helm-ai-kernel boundary status --json
helm-ai-kernel receipts tail --agent agent.demo.exec --server http://127.0.0.1:7714
helm-ai-kernel verify evidence-pack.tar
Feedback question: Is there a factor in 12-factor-agents that explicitly covers audit trail integrity / tamper-evident logging? If not, do you see it as a gap worth addressing in a future revision?
Repo: https://github.com/Mindburn-Labs/helm-ai-kernel
Hi — 12-factor-agents is a great framework for thinking rigorously about what makes agents production-ready. I wanted to raise a pattern that might be worth adding or referencing: tamper-evident decision audit.
The 12 factors cover a lot of great ground — statelessness, human-in-the-loop, tool boundaries, etc. One pattern that seems missing is: how do you prove to an auditor what an agent actually decided and executed, after the session ends, without relying on mutable log files?
We built HELM AI Kernel to address exactly this. For every ALLOW/DENY/ESCALATE decision, HELM produces a cryptographically signed receipt and bundles them into an offline-verifiable EvidencePack — a tamper-evident bundle of what ran, when, and under which policy.
This matters for:
Quick test (~5 min):
Feedback question: Is there a factor in 12-factor-agents that explicitly covers audit trail integrity / tamper-evident logging? If not, do you see it as a gap worth addressing in a future revision?
Repo: https://github.com/Mindburn-Labs/helm-ai-kernel