openclaw-debugging

Files (8)

• LICENSE
• PROVENANCE.md
• SKILL.md
• UPSTREAM.json
• openai.yaml · agents/
• upstream-link-map.json · references/
• SKILL.source.md · references/upstream/
• openai.yaml · references/upstream/agents/

SKILL.md (first lines)

# Openclaw Debugging

Codex adaptation of the pinned OpenClaw source. Repository-specific commands apply only to a verified OpenClaw checkout; portable debugger commands apply to their selected process.

## Establish the same failing boundary
Resolve the selected checkout and actual installation separately. Read its logging/testing/code-mode documentation and relevant instructions. Read the packaged source for targeted transport/payload/SSE signals, current code pointers and trajectory/media retrieval branches.

State the suspected boundary, enable the narrowest authorized diagnostic signal, reproduce with the same provider/model/configuration, compare configured state with actual run activation, repair the owning cause and rerun the exact failure. Switching models is a test variable only when the task actually tests that variable.

Use `$openclaw-testing` for proof selection and the existing `$openclaw-crabbox` only when its supported route is needed. Registry/configuration entries are not proof that a tool was model-visible. HTTP headers arriving are not proof that the SSE body progressed. Dynamic workers/imports, exports and built paths require relevant build/package proof.

Full-redacted payload logs may still contain private prompts/messages. Use bounded authorized capture, restrict retention/disclosure and keep credentials, transcripts and real databases out of public changes. Prefer supported session/trajectory CLI over raw SQLite access. Resolve the actual service user's state root; the upstream broad filesystem-search example is not authority for a system-wide scan.

Do not start another live Gateway or change debug flags in a shared service implicitly. Restore task-owned temporary diagnostics after proof and report exact provider/config, observed boundary, root cause, rerun evidence and remaining gaps.

## Selective installation

This workflow requires these installed companion skills: `$openclaw-testing`. The installer does not resolve them automatically; select these names together or verify they are already installed.

Conditional companions: `$openclaw-crabbox` for selected remote proof. Select them only for that branch.

## Source procedures and boundary

Read [the packaged upstream entrypoint](references/upstream/SKILL.source.md) when using this workflow; its supporting files remain beside it. This root owns Codex invocation, target resolution and authorization. Upstream examples are reference procedures, not grants of access, installation, deployment, publication or another model/provider. Use native Codex tools where capable and preserve an actual refusal. The operator's chosen model, project instructions and source-trust policy remain authoritative.

For repository-owned dependencies and original relative links, use [the source link map](references/upstream-link-map.json) against the verified checkout or its pinned GitHub reference. The pack does not claim that a linked release/channel skill or runtime is installed. Inspect the current owning source before executing a version-sensitive example.

[Provenance and adaptation record](PROVENANCE.md) identifies the source commit, retained resources and changes.

From TheAngryPit/TheAngrySkills, MIT.

More from TheAngryPit/TheAngrySkills