Back to Ecosystem Pulse
EvaluationExperimental
self-care
by Not-Diamond
Trace-based detection of agent failures and context remediation
JavaScript
Updated Apr 16, 2026
Share:
Summary
Analyzes agent execution traces from Claude Code to detect quality issues like goal drift, hallucinations, and missed actions. Works as a plugin that inspects trace events and context snapshots to flag failures and suggest remediation points. Exposes actionable diagnostics and integrates with OpenTelemetry-style traces for provenance and continuous monitoring continuous monitoring.
The Value Proposition
As agents become more autonomous, root-causing where and why they fail is essential for building trust and improving agent-to-agent interactions agent-to-agent interactions. This tool surfaces concrete failure modes in execution traces so teams can measure agent reliability and iterate on remediation. By turning traces into reproducible failure signals, it helps move evaluation from one-off benchmarks to continuous agent track record analysis.
Best For
Teams running Claude Code agents who need automated detection of hallucinations Hallucination Propagation, goal drift Context Drift, and missed actions in execution traces.
Real-World Examples
- Detecting hallucinations and goal drift in multi-step agent traces
- When you need automated remediation hints from execution context snapshots
- Integrating trace-based diagnostics into a continuous agent evaluation pipeline
Works With
claude-codeopentelemetry
Topics
agent-evaluationai-agentsclaude-codellmobservabilityopentelemetryplugintracing
Similar Tools
agent-playground
Keywords
A2A evaluationagent track recordagent failure modesagent-evaluation