Agents Shipgate provides a deterministic merge gate that evaluates whether AI‑generated agent capability changes are safe to merge. By performing static, local‑first analysis—without executing agents, making tool calls, or contacting LLMs—it verifies tool‑use readiness, such as newly added Stripe refund APIs, before code integration. This helps developers enforce release policies and avoid runtime failures in AI‑driven workflows.
Funding
Funding not disclosed
Founders
Product
Problem
When a pull request modifies the set of tools an AI agent can use—such as adding a new payment or email capability—teams lack a deterministic, pre‑merge check to confirm that the change complies with security, policy, and idempotency requirements.
Solution
Agents Shipgate provides a static, local‑first CLI and GitHub Action that analyzes the declared tool surface in a repository’s shipgate.yaml manifest and associated tool specifications. By performing a deterministic merge gate analysis, it determines whether the capability change can be safely merged without executing any agents, invoking LLMs, or making network calls. The tool produces a merge verdict, detailed JSON and SARIF evidence, and optional PR comments for human reviewers. This enables release owners to enforce policies such as approval, idempotency, and blast‑radius constraints before code lands. The solution integrates into existing CI/CD pipelines, offering advisory, baseline, and strict modes to suit different governance maturity levels.
Target Audience
Primary users are engineering teams that develop and deploy tool‑using AI agents, particularly those using GitHub‑based CI/CD workflows and requiring pre‑deployment security and policy checks.
Features
- Open‑source CLI and GitHub Action named agents‑shipgate that runs entirely offline
- Parses checked‑in shipgate.yaml manifests, OpenAPI specs, and SDK metadata to compute tool‑surface diffs
- Generates deterministic merge verdicts (pass, block, review) with evidence in JSON and SARIF formats
- Posts automated PR comments and can enforce merge protection rules in CI pipelines
- Supports multiple verification modes (advisory, baseline, strict) for progressive policy enforcement
- No execution of agents, tool calls, LLM invocations, or external network requests by default