| name | local-agentops |
| description | Analyzes AI Agent CSV/JSON logs locally and generates grounded operations reports. Invoke for Agent evaluation, failure diagnosis, version comparison, latency, cost, or handoff review. |
Local AgentOps
Use the repository-owned implementation in skills/local-agentops/.
Invoke
Run the fixed entry point for the current platform:
skills\local-agentops\scripts\run.ps1 "<input.csv|input.json>" --output "report.md" --json "metrics.json"
skills/local-agentops/scripts/run.sh "<input.csv|input.json>" --output "report.md" --json "metrics.json"
For a deterministic smoke test without model inference:
skills\local-agentops\scripts\run.ps1 "examples\agentops-sample.csv" --no-model
Required Behavior
- Keep the input file on the local machine.
- Report valid and skipped records before interpreting results.
- Treat deterministic metrics and threshold rules as the primary evidence.
- Treat the OpenVINO model output as an auxiliary summary only.
- Never fall back to a cloud model.
- Do not claim causality or statistical significance.
- Return the generated Markdown and optional JSON report paths.
Expected Inputs
Required result field: success, passed, or status.
Optional fields:
task_id, trace_id, or run_id;
latency_ms or duration_ms;
cost_usd or cost;
human_review or human_handoff;
error_type or error_category;
prompt_version, agent_version, or version;
user_rating or rating.
Read skills/local-agentops/SKILL.md for the full data contract, failure handling,
privacy boundary, and examples.