agentcompass summary

Regenerate benchmark aggregates and summary.md from an existing run directory.

agentcompass summary reads existing task details, recomputes the benchmark’s aggregate metrics, and regenerates summary.md:

agentcompass summary [OPTIONS] RUN-DIR

It does not rerun agents, verifiers, or analyzers.

Regenerate a Summary

agentcompass summary \
  results/swebench_verified/$MODEL_NAME/20260703_120000

The run directory must contain the persisted AgentCompass metadata and details/ needed to reconstruct the benchmark result.

Preview Without Writing

Use --dry-run to print the regenerated Markdown without replacing summary.md:

agentcompass summary \
  results/swebench_verified/$MODEL_NAME/20260703_120000 \
  --dry-run

Options

Option	Purpose
RUN-DIR / --run-dir	Existing AgentCompass run directory. Required.
--config <path>	Load an additional configuration override; repeatable.
--benchmark-params <json>	Override benchmark fields needed while rebuilding aggregates.
--dry-run	Print the regenerated summary.md without writing files.

Use benchmark overrides only when the benchmark’s aggregation logic genuinely requires them. Changing measured task settings after execution does not make the stored results comparable to a differently configured run.

Choose Summary or Analysis

Need	Command
Detect bad cases, calculate trajectory statistics, or run qualitative diagnosis	agentcompass analysis
Recompute benchmark totals and metrics from existing details	agentcompass summary
Execute missing or failed benchmark tasks again	agentcompass run with deliberate reuse

See Results for artifact roles and persisted run data.