Operations · Advanced
Verified
Automated verification
- Passed every automated gate: schema, uniqueness, HTTPS and live source link, prompt-injection and safety scan, approval-step rule.
- Drafted by google/gemini-3.8-flash via OpenRouter.
- Second-model review: x-ai/grok-4.3 (no critical or high findings).
- Verified .
- Automated checks and AI review only, not human review.
Auditing agent execution traces for silent failures
Inspect multi-step agent runs to detect tool errors, bad reasoning, and failure patterns.
- On a scheduleHourly trace batch review
- ThenAnalyze spans and tool calls
- ResultDraft issue with repro trace
Review full execution logs and spans from multi-step autonomous workflows to pinpoint silent failures where tools or models deviated from intended goals. Flag anomalies, cluster repeated failure modes, and format reproducible test cases for development teams to resolve.
Prompt to start with
BotBentoRecommended
Analyze incoming execution traces stored in the database. Inspect each span, prompt, and tool call to verify whether the agent achieved its objective without silent errors or incorrect tool arguments. Group identical failure patterns into common categories and compile a debugging report containing representative traces. Before creating a GitHub issue or sending any notifications to Slack, display the findings and wait for user approval.