01
Look past the change
We show which files run later (a workflow, an install script, a hook or instructions for another agent), when they run and with which declared permissions.
For AI teams and developers
SecureStamp surfaces workflows, scripts and instructions a change can set up to run, along with their declared permissions. Development and security review the same evidence and tie approval to the exact artifact.

Execution Governance sets the controls that define which effects an agent may cause, on which resources, and what evidence remains.
01
We show which files run later (a workflow, an install script, a hook or instructions for another agent), when they run and with which declared permissions.
02
Whatever could not be observed is reported as unknown, never as safe. When there is a risk, we suggest the fix (for example, making a workflow read-only) without applying it for you.
03
Approval stays tied to the exact change that was reviewed, and the evidence comes out as a bundle your team can verify offline, without relying on us.
These checks follow SAES, the public standard for structuring tasks, authority, safeguards and evidence when delegating work to agents. Explore SAES
Development arrives at review with the evidence assembled. Security authorizes a concrete change, not open-ended access. And both argue over the same facts.
You asked for a calculation fix. The agent fixed it and, along the way, touched a workflow that will run tomorrow with other permissions. The diff was small; what it set up was not. That is why the first thing you see is what the change enables, before anyone says yes.
The report shows what was evaluated, which permissions are declared and which paths were outside observation. With that scope in view, the team can approve or request an additional review.
Install the published package
Install the local MCP wrapper to keep a tool call inside the controls you already use.
npx -y @securestamp/mcp-guard@0.2.1Open the lab
Use the lab to test declared limits across MCP, shell, scripts and APIs before granting autonomy.
Open the labAgent reports
Review agent and MCP client activity observed by SecureStamp, the findings it surfaced, and when human confirmation was needed.
Every finding counts actions that went through SecureStamp and links to the evidence. The report does not infer agents' motives or read their reasoning.
The file carries a SecureStamp ES256 signature. Anyone can check it with a POST to /api/agents/reports/verify or against the public JWKS.
Shows a banner on Agents and MCP Guard. It never blocks or allows anything.
A small, explicit quota for trying the protected flow, with limits visible before you connect an agent. Paying changes how much you can use, not the conclusion: the same evidence gets the same evaluation on every plan.