Observed arrival · 2026-09-04
tracelint reads the agent trace, not the alibi
A deterministic linter for tool-calling agent traces that flags structurally provable failures and can fail CI on them.
Field notes
The tool normalizes spans from systems such as Langfuse, Phoenix, LangSmith, OpenAI, and OpenTelemetry before applying deterministic structural checks. Its worked example records an HTTP 500 from get_order, reuses the failed order value in refund_order, and invokes that side-effecting operation twice. The page says findings can be emitted as JSON or SARIF and used to fail CI, while emphasizing that the checks do not call a model or judge the final prose.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue