agentcompass summary

agentcompass summary recomputes benchmark aggregate metrics from task results in an existing run directory and updates the summary files. It does not rerun agents, verifiers, or analyzers.

agentcompass summary [OPTIONS] RUN-DIR

Regenerate a Summary

RUN-DIR should point to the complete AgentCompass result directory, not its details/ subdirectory:

agentcompass summary \
  results/swebench_verified/my-model/20260703_120000

AgentCompass reads the saved task results and recomputes metrics with the corresponding benchmark’s aggregation logic. By default, it creates or replaces summary.md and the internal .summary_counts.json file in place without modifying task results.

Preview the Summary

Use --dry-run to preview the new summary.md in the terminal before writing files:

agentcompass summary \
  results/swebench_verified/my-model/20260703_120000 \
  --dry-run

This mode does not update any files in the run directory.

Parameters

Parameter	Required?	Default	Purpose
RUN-DIR / --run-dir	Required	None	Selects an existing AgentCompass run directory.
--config <path>	Optional	None	Loads a YAML or JSON file that supplements benchmark configuration; repeatable. Persisted parameters with the same keys take precedence.
--benchmark-params <json>	Optional	None	Uses a JSON object to override benchmark parameters recovered from the run directory.
--dry-run	Optional	Disabled	Prints the regenerated Markdown to the terminal without writing files.

You normally do not need --config or --benchmark-params. Use them only when an older run lacks fields needed for aggregation or when you intentionally want different aggregation parameters. They affect only this summary operation and do not reevaluate tasks.

Related Pages

• Summary and Analysis Results
• agentcompass analysis
• agentcompass config
• CLI Overview