Problem Description
The audit tool exceeds its expected <30s performance target when auditing long-running or complex workflows. Performance degrades significantly with workflow duration and turn count.
Command/Tool
- Tool:
audit
- Command:
agenticworkflows audit --run_id <run_id> --max_tokens 3000
Steps to Reproduce
- Use the
logs tool to find a long-running workflow run (>8 min duration)
- Call
audit on that run ID
- Observe elapsed time in the
Tool call complete log line
Expected Behavior
Audit should complete in under 30 seconds for most workflow runs, per documented performance targets.
Actual Behavior
Observed during daily exploratory testing (run §28079720278):
| Run ID |
Workflow |
Duration |
Turns |
Audit Time |
| 28079043748 |
Contribution Check |
5.7m |
~0 |
12.6s ✅ |
| 28079516670 |
PR Sous Chef |
8.6m |
~0 |
~52s ❌ |
| 28078517319 |
Static Analysis Report |
12.2m |
32 |
72.3s ❌ |
The audit for Static Analysis Report (Claude engine, 32 turns, 12.2 min) took 72.3 seconds — 2.4× over target.
Environment
- Repository: github/gh-aw
- Tester Run ID: 28079720278
- Date: 2026-06-24
Impact
- Severity: Medium
- Frequency: Always (scales with workflow complexity/duration)
- Workaround: None — user must wait
Root Cause Hypothesis
Audit time appears to scale with workflow log volume. For workflows with many turns or long durations, the log download and parsing step likely dominates.
Additional Context
The compile tool (251 workflows in 20.9s) and basic audit for short runs (~12.6s) are well within targets. The performance issue is specific to complex, long-running runs.
Generated by 🧪 Daily CLI Tools Exploratory Tester · 95.7 AIC · ⌖ 17.7 AIC · ⊞ 9.6K · ◷
Problem Description
The
audittool exceeds its expected <30s performance target when auditing long-running or complex workflows. Performance degrades significantly with workflow duration and turn count.Command/Tool
auditagenticworkflows audit --run_id <run_id> --max_tokens 3000Steps to Reproduce
logstool to find a long-running workflow run (>8 min duration)auditon that run IDTool call completelog lineExpected Behavior
Audit should complete in under 30 seconds for most workflow runs, per documented performance targets.
Actual Behavior
Observed during daily exploratory testing (run §28079720278):
The audit for Static Analysis Report (Claude engine, 32 turns, 12.2 min) took 72.3 seconds — 2.4× over target.
Environment
Impact
Root Cause Hypothesis
Audit time appears to scale with workflow log volume. For workflows with many turns or long durations, the log download and parsing step likely dominates.
Additional Context
The
compiletool (251 workflows in 20.9s) and basicauditfor short runs (~12.6s) are well within targets. The performance issue is specific to complex, long-running runs.