| name | audithub-orca-call-metrics-analyzer |
| description | Analyze OrCa call_metrics.json outputs for fuzzing efficiency and blockers. Use when a user provides call_metrics.json, asks why functions revert or have low successful-call percentages, wants hint usage analysis, needs per-function metrics triage, or wants tuning hypotheses for an OrCa campaign. |
OrCa call metrics analyzer
Overview
Turn OrCa metrics into a prioritized explanation of what the fuzzer can and cannot currently exercise.
Required reference
Read ../references/call-metrics-and-tuning.md.
Sub-agent use
This skill is a strong default candidate for sub-agent execution because call_metrics.json can be large. When used as a sub-agent, return normalized analysis instead of raw metrics:
- top blockers and strategic functions
- dominant revert reasons/selectors and suspected source checks
- hint usage or hint failure issues
- functions that look healthy enough to leave alone
- concrete tuning hypotheses with expected validation metrics
For later campaign stages, save the full raw metrics to orca_config/analysis/call_metrics_<run>.json or another local artifact, then pass only this summary back to the orchestrator. If comparing runs, read both metrics files locally and return a compact diff of changed success rates, new blockers, resolved blockers, and hint effectiveness.
Workflow
- Parse
call_metrics.json; do not rely on visual inspection when structured parsing is available.
- Rank functions by low success rate, high revert volume, and strategic importance.
- Review dominant revert messages/selectors and map them to source checks.
- Identify likely blockers: arguments, access, balances, approvals, state ordering, or bad setup.
- Review hint usage and hint failures when hints exist.
- Store normalized analysis in
orca_config/analysis/call_metrics_<run>.json when writing artifacts.
Output
Report:
- top poorly performing functions and success rates
- dominant revert reasons and suspected source checks
- functions that are acceptable as-is and why
- hint usage issues
- concrete tuning hypotheses for
audithub-orca-setup-tuner
Treat low success as a signal for investigation, not an automatic defect. Preserve fuzzing breadth in recommendations.
