Deep Dive
Runtime execution enforcement for AI coding.
Managing the execution boundary of AI coding tools through policy, attribution, and evidence. If an action is not approved by policy, it does not execute.
Approval Gating
Principle 01Every consequential action requires an explicit operator checkpoint before side effects can run.
Approval gating ensures that every consequential AI-initiated action requires explicit human authorization before execution. Without approval gating, AI tools can execute file modifications, shell commands, and repository pushes autonomously. Approval gating closes the gap between what an AI proposes and what actually executes, with binding to exact action arguments, not natural language descriptions.
Mechanism: Argument Binding
Every consequential action requires explicit human authorization before execution. Approvals bind to SHA-256 digests of exact action arguments, not natural language descriptions. If arguments change after approval, execution is denied.
View Proof Artifacts →Attributable Audit
Principle 02Append-only evidence records that survive review and show who approved what, when, and under which policy.
An audit trail for AI coding tools records who approved an action, what exact arguments were approved, what executed, and the outcome. Syndicate Code's audit trail is append-only and cryptographically chained (SHA-256 hash chain from the genesis event), providing tamper-evident provenance for every approval, denial, and execution. This enables deterministic replay and post-incident attribution.
Mechanism: Cryptographic Chain
Audit events are stored in a local SQLite database with Write-Ahead Logging (WAL) and are HMAC-signed using an application secret. Every event contains the hash of the preceding event, creating an immutable record of execution history.
View Proof Artifacts →What this page does not claim
- It does not claim universal security or kernel-level isolation.
- It does not claim to prevent prompt injection in the model itself.
- It does not claim to integrate with external AI coding tools as an enforcement layer.
- It does not claim that enforcement is preserved when the control plane is unavailable.
Frequently Asked Questions
What is runtime enforcement for AI coding?
Runtime enforcement is the policy, approval, and evidence control path that decides whether an AI-assisted action can execute and which human-approved record is written for it.
Why does my team need approval gating?
Approval gating requires an operator to review the exact action arguments before execution. That prevents silent drift between what the AI proposed and what actually ran.
How does Syndicate Code differ from observability?
Observability records what happened after the fact. Syndicate Code evaluates policy before execution and can block a disallowed action at the execution boundary.