openclaw-debugging

Diagnose OpenClaw model, provider, tool-surface, and streaming issues using targeted logs and probes.

1|Updated Jun 9, 2026
One-click install
npx skills add https://github.com/JEROME-PRAKASH-L/openclaw --skill openclaw-debugging-jerome-prakash-l
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: openclaw-debugging
Source: https://github.com/JEROME-PRAKASH-L/openclaw/tree/main/.agents/skills/openclaw-debugging
Command: npx skills add https://github.com/JEROME-PRAKASH-L/openclaw --skill openclaw-debugging-jerome-prakash-l

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When OpenClaw behaves differently across local tests, live models, providers, code mode, Tool Search, Crabbox, or CI, it is hard to know where to look. This Skill replaces guessing with a structured debug loop that picks the right logs, probes, and proof path before any code is changed. ## Core Features & Use Cases - Boundary-first debugging: Identify whether the fault lies in config, tool construction, provider payload, fetch, SSE streaming, transcript replay, worker/runtime, packaging, or CI before patching. - Targeted transport logging: Use narrow env flags like OPENCLAW_DEBUG_MODEL_TRANSPORT, OPENCLAW_DEBUG_MODEL_PAYLOAD, OPENCLAW_DEBUG_SSE, and OPENCLAW_DEBUG_CODE_MODE instead of noisy global debug output. - Proof selection guidance: Match the fix to the right verification lane, from targeted Vitest runs and pnpm build for worker/dist changes to Crabbox for Docker, Linux, and CI-parity proof. - Use Case: A configured tool never reaches the model in a live session. Use this Skill to compare configured state against actual run activation, inspect the final model-visible tool names, and confirm whether code mode reduced the surface to exec and wait. ## Quick Start Ask the assistant to use the openclaw-debugging skill to find why a configured tool is not visible to the live model and which debug flags to enable.

Frequently Asked Questions about openclaw-debugging

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

FAQPage Schema
How do I debug OpenClaw model streaming issues?

Enable OPENCLAW_DEBUG_MODEL_TRANSPORT=1 to log request start, fetch response, first SSE event, and stream completion. Add OPENCLAW_DEBUG_SSE=events or peek to inspect stream timing and the first redacted events, then watch output with openclaw logs --follow.

How do I check which tools the model actually sees in OpenClaw?

Set OPENCLAW_DEBUG_MODEL_PAYLOAD=tools to log all model-facing tool names before the SDK call. Compare configured state against actual run activation, since config can be enabled while the run disables tools or code mode reduces the surface to exec and wait.

Why is a configured OpenClaw tool not active at runtime?

Config can be enabled while the run is raw, has an empty allowlist, or lacks model tool support. Inspect the final model-visible tool names rather than the registry, and check whether code mode actually activated before enforcing payload invariants.

What is the difference between fetch and SSE debug signals?

A fetch response proves HTTP headers arrived, while the first SSE event proves the provider body is progressing. A gap between them indicates a stream, body, or provider issue rather than a tool execution problem.

When should I use Crabbox instead of local tests for OpenClaw debugging?

Use Crabbox for broad, Docker, package, Linux, live-key, or CI-parity proof. Local targeted Vitest suits single helper or payload bugs, while worker, dist, or lazy-import changes need targeted tests plus pnpm build.