Trajectly

Trajectly provides deterministic regression testing for agents, ensuring their consistent and reliable operation. Key capabilities include: * Baseline recording of agent behavior * Enforcement of behavioral contracts and rules * Deterministic replay for consistent results * Precise identification of divergence and regression points * CI-native integration for continuous validation This robust testing framework addresses common challenges in agent development, such as silent regressions and flaky output comparisons. Instead of subjective raw output diffs, Trajectly employs behavioral refinement to verify the preservation of core sequences and tool calls. It allows developers to define and enforce explicit contracts covering tool usage, call sequences, budget thresholds, and data safety, ensuring agents adhere to predefined operational boundaries. The system provides exact witness reporting for any failure, pinpointing the specific event of divergence with a violation code and a repro command for easy debugging. Trajectly works by first recording an agent's normal execution to capture traces, then deterministically replaying these traces using fixtures to bypass external dependencies like live API keys in CI environments. It then compares the replayed trace against the baseline, applying behavioral refinement and contract checks to deliver a clear PASS or FAIL verdict. This approach is ideal for ensuring the stability and predictability of complex agents. This tool is designed for engineering teams and developers who build and deliver agents, especially those operating in critical environments where behavioral consistency and reliability are paramount. It integrates seamlessly into continuous integration pipelines, allowing teams to catch regressions early and maintain high quality throughout the development lifecycle.