halo-loop

Diagnose HALO trace data to surface harness failure modes with evidence.

1.1k|87|Updated Apr 21, 2026

Search Tags:#opentelemetry #diagnostic #harness #rlm #trace-analysis #halo

Ready to run

npx skills add https://github.com/context-labs/HALO --skill halo-loop

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: halo-loop
Source: https://github.com/context-labs/HALO/tree/main/skills/claude
Command: npx skills add https://github.com/context-labs/HALO --skill halo-loop

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

HALO Loop provides a diagnostic-guided harness improvement workflow. It analyzes HALO traces to surface failure modes and enables Claude to map findings to code, apply minimal changes, re-run, and measure impact.

Core Features & Use Cases

Diagnostic surface: identifies common harness failure modes from HALO trace data.
Code-integration mapping: translates findings into small, verifiable code changes and re-runs HALO.
Measured iteration: compares before/after traces to quantify improvement in harness reliability.

Quick Start

Provide a path to traces.jsonl and a diagnostic prompt for HALO to analyze so HALO returns evidence-backed findings.

Frequently Asked Questions about halo-loop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I diagnose agent harness failure modes from OpenTelemetry trace data?▼

HALO Loop analyzes OpenTelemetry trace data to surface common harness failure modes with trace evidence. It identifies issues in harness prompts, tool configurations, and retry logic to guide targeted improvements.

What's the best way to iteratively improve agent harness reliability using trace analysis?▼

Iterative harness improvement involves applying minimal code changes based on diagnostic findings, re-running HALO, and measuring before/after traces. This cycle quantifies reliability gains across each iteration.

How does diagnostic-guided harness improvement work for agent failures?▼

Diagnostic-guided improvement maps HALO trace findings to small, verifiable code changes in harness prompts or tool configurations. After applying changes, it re-runs HALO to measure the impact on reliability.

Can I fix agent harness retry logic and tool configurations without changing the model?▼

Yes, HALO Loop focuses on harness-level fixes rather than model-level changes. It targets harness prompts, tool configurations, and retry logic to resolve failure modes identified in trace data.

How do I start analyzing agent traces to find harness failure modes?▼

Provide a path to a traces.jsonl file and a diagnostic prompt. HALO analyzes the trace data and returns evidence-backed findings that surface common harness failure modes for actionable code changes.

When should I not use trace analysis for harness improvement?▼

Trace analysis for harness improvement targets harness prompts and tool configurations, not model-level fixes. If failures stem from underlying model capabilities, this diagnostic approach will not resolve them.