$ loading_
Validate AI agent behavior contracts and detect regressions through baseline comparisons.
Copy the install command and let the AI configure it · recommended for beginners
No copy-paste install info for "agent-invariants" yet — see the docs or source repo.
Using this normalized event trace, check the AI agent against the operating contracts for stop, approval, scope, and completion rules, and list any violations.
A contract validation report showing which rules passed, which failed, and the related anomalies.
Compare the normalized event traces of the baseline and candidate versions, focusing on regressions in approval, recovery, stop, and completion behavior, and summarize the differences.
A regression analysis highlighting behavioral degradations in the candidate relative to the baseline, with a summary of differences.
Given the normalized event trace of an agent in an error scenario, check whether its recovery flow follows the defined contracts and identify any incomplete or out-of-scope actions.
A contract-check result for the recovery flow, indicating whether recovery was handled correctly and where issues may exist.
Developers or researchers can compare baseline and candidate event traces after updating an agent policy to catch regressions early. It is especially useful for checking whether key rules such as approval, stop, and completion have been broken.
When a team needs stable operating contracts for an AI agent, this tool can perform deterministic checks on normalized event traces. It helps turn behavior acceptance from subjective observation into a repeatable validation process.
When an agent handles failures inconsistently, teams can use it to review whether recovery and scope rules were followed. This helps identify recovery failures, out-of-scope actions, or incomplete flows more quickly.
It is a deterministic behavior-compatibility layer for AI agents. It checks whether normalized event traces satisfy operating contracts and compares baselines with candidates to find regressions.
Based on the description, it is designed to catch degradations or violations in approval, stop, scope, recovery, and completion rules. In other words, it focuses on whether agent behavior deviates from defined contracts.
The provided materials do not include installation steps, runtime requirements, or key information. For exact prerequisites and setup details, see the source repository.
Compare AI-agent traces and enforce declared behavior contracts.
Add offline, deterministic pre-action safety gates with signed verdicts for AI agents.
Verify AI agent outputs with reviews, signed proofs, and public ledger records.
Profile AI agents locally to trace costs, behavior issues, and config fixes.
Run safety-invariant checks on structured outputs to protect locked fields.
Validate agent outputs, detect patterns, and solve mathematical problems deterministically.