Before accepting a change
Verify the recorded run against the pinned repository commit, exact command, and required checks before accepting an agent-produced change.
Downloadable B2B software · Version 1.0.0
Turn agent-run evidence into a deterministic pass or fail.
Agent Eval Kit is a standalone Python application that validates a versioned task contract, seals it with SHA-256, and evaluates recorded evidence against the pinned repository commit, exact command, and required checks. It runs offline and has no third-party runtime dependencies.
The caller runs tests in its own isolated environment; Agent Eval Kit validates only recorded JSON evidence and never executes customer code.
No network access
No arbitrary code execution
No telemetry
When it fits
Use it as the final acceptance gate.
Use Agent Eval Kit after the caller has run its own isolated tests and before accepting an agent-produced change or handing its evidence to another team.
At a team handoff
Re-run the same JSON contract and evidence offline so a release or evaluation team can independently reproduce the decision.
When evidence changes
Reject a changed commit, mismatched command, or missing required check with a nonzero exit and zero reward—no partial credit.
Company license
Agent Eval Kit Team License
USD 1,000
One-time purchase · 25 internal users · Versioned digital archive.
- Perpetual internal-use license for one legal entity
- Standalone Python zip application
- Python 3.11 or newer; no third-party packages
- Contract validation and canonical SHA-256 sealing
- Fail-closed evidence evaluation with no partial credit
- Examples, product documentation, and checksum
Checkout is hosted by PayPal. Delivery is completed to the payer email within 24 hours after cleared payment and confirmation of the licensed legal entity; no account, meeting, or shipping is required. Email support is limited to installation questions and reproducible defects in the purchased version.
Under the Refund Policy, a full refund may be requested within 14 calendar days if the archive has not been delivered or accessed, with a documented remedy and refund path for material reproducible defects.
Release evidence
A real, reproducible software artifact is ready.
The current archive is agent-eval-kit-1.0.0.pyz. Its SHA-256 is 71580501a6004ae63e2443a5b8bac61dd84411b3dccdd5ad532f002e45e515d7. Twelve automated tests cover contract validation, tamper rejection, exact check schemas, structured failures, standalone execution, and byte-for-byte reproducible builds.
Product boundary
Software only, with a fixed binary and fixed license.
Included
The versioned application, examples, documentation, checksum, and the purchased license grant.
Not included
No consulting, No custom development, no managed evaluation, and no access to customer systems.
Delivery
Within 24 hours after the payment processor confirms a completed, cleared payment and the licensed legal entity is confirmed by email, the Supplier sends the unmodified archive, published checksum, and license record to the payer email address.