Description
signed-audit-trails-recipe is an agent skill from the wshobson/agents collection that is a step-by-step cookbook for setting up cryptographically signed audit trails on Claude Code tool calls. It is the teaching companion to the protect-mcp plugin, which provides the runtime implementation. The idea is that every tool call, such as a Bash command, an Edit, a Write or a WebFetch, is evaluated against a Cedar policy before it runs, so a call the policy denies does not execute, and is then recorded as an Ed25519-signed receipt after it runs. The receipts use a canonical JSON form and can be verified offline by anyone holding the public key, so an auditor or counterparty can check the record later without contacting the operator. The skill walks through what this gives you, when the pattern is worth using, and the steps to set it up and verify the receipts. It is intended for teams that need evidence of what an agent did, for example in regulated work, and for people evaluating or demonstrating the approach before adopting it. The recipe shows the mechanism, and the runtime setup is handled by the plugin the recipe refers to. The skill is installed with the skills CLI from the wshobson/agents repository. Its skills.sh page showed about 5.9K installs and passing results from three security audits at the time of the scan.…
Highlights
- Step-by-step cookbook for signed audit trails
- Cedar policy check before each tool call
- Ed25519-signed receipts after each call
- Offline verification by an auditor
- Companion to the protect-mcp plugin
- Installs with the skills CLI
Ratings & Reviews
No ratings yet
Be the first to rate signed-audit-trails-recipe.
People also visit
review-agent-setup
Configure human-in-the-loop gating for AI agent review actions in Claude Code
signed-audit-trails-recipe FAQ
What does this recipe teach?
How to set up signed audit trails for Claude Code tool calls, with policy checks before each call and signed receipts after.
Does a denied call still run?
No. If the Cedar policy denies a call, the tool does not run.
Who can verify the receipts?
Anyone with the public key can verify them offline, without a network call or trust in the operator.
How is the runtime installed?
Through the protect-mcp plugin, which the recipe refers to for the implementation.
How do I install it?
Run npx skills add https://github.com/wshobson/agents --skill signed-audit-trails-recipe.
What security audits does it have?
skills.sh lists three audits: Gen Agent Trust Hub Pass, Socket Pass and Snyk Pass.
What are the best signed-audit-trails-recipe alternatives?
Popular alternatives to signed-audit-trails-recipe include protect-mcp-setup, agent-governance and block-no-verify-hook. See all signed-audit-trails-recipe alternatives