Export a local evidence report from coding-agent sessions
At a glance
AgentTrace's discussion documents a local inspection workflow for spotting tool failures, delays, and cost patterns in saved agent sessions. This workflow is listed under Agent Debugging, Session Review, Observability.
- Agents / tools
- AgentTrace, Local coding-agent logs
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Have saved session logs from a supported coding agent on a machine with Homebrew. Install AgentTrace using a documented route, such as brew install luoyuctl/tap/agenttrace. 2) Run agenttrace --doctor to check the local setup, then launch agenttrace to inspect sessions in its terminal interface. 3) Generate machine-readable evidence with agenttrace --overview -f json. Examine failures, repeated calls, long gaps, token usage, and available cost fields. 4) Save a shareable report with agenttrace --overview -f html -o agenttrace-overview.html and open that file to review the evidence. 5) For a recurring workflow, run agenttrace --overview --fail-under-health 80 --fail-on-critical --max-tool-fail-rate 15 to apply the discussion's example health gate and inspect flagged runs.
Worth knowing
The discussion distinguishes unmatched calls from explicit timeouts: an absent result leaves completion unknown. Its maintainer reports correcting that distinction; a missing result does not establish whether a remote action succeeded.