Verify signed agent-run evidence offline with Tracekit
Check an authorized Tracekit bundle against pinned signer keys and report integrity verdicts separately from execution and completeness claims.
npx skills add agentskillexchange/skills --skill verify-signed-agent-run-evidence-offline-with-tracekit
Invoke when an operator receives an agent-run evidence bundle and needs an independently checkable integrity result rather than a visual trace review. The agent identifies the bundle and trusted signer key, runs offline verification and returns the verdict with its limits.
What this skill actually does
Install the upstream Python package tracekit-ai; the unrelated tracekit PyPI package is not this project. With a trusted checkout, validate the documented fixtures using tracekit verify docs/sample/demo-run.tkb –key docs/sample/signer.pub and the corresponding demo-run-tampered.tkb command. Upstream documents success for the intact fixture and failure for the modified one.
For an authorized real bundle, use tracekit verify –key . Establish trust in the signer key separately from the submitted bundle. Record the verifier version, command, exit status, verdict and any identified failing record. Retain the original evidence and return unresolved format or key issues explicitly.
The boundary is signed-bundle integrity verification, not general observability or agent benchmarking. Verification cannot establish truthful tool results, complete action capture or intent. Same-user development signing offers no isolation from that user; truncation and signer compromise require additional witness arrangements. Sources: https://github.com/Cygnux-Labs/Tracekit and its upstream sample-bundle instructions.