debug-inference

Debug why inference.local, direct external inference, or supervisor-only system inference is failing. Use when the user cannot reach a local model server, has provider base URL issues, sees inference verification failures, hits protocol mismatches, or needs to diagnose inference on local vs remote gateways. Trigger keywords - debug inference, inference.local, system inference, sandbox-system, local inference, ollama, vllm, sglang, trtllm, NIM, inference failing, model server unreachable, failed to verify inference endpoint, host.openshell.internal.

Quality

83%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Medium

Suggest reviewing before use

Quality

Content

77%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A highly actionable, well-sequenced diagnostic workflow with concrete commands and strong validation/feedback loops. Its main weaknesses are redundant content between the workflow, failure-pattern table, and diagnostic dump, and a monolithic single-file structure with no progressive disclosure into reference files.

Suggestions

De-duplicate the Common Failure Patterns table and Full Diagnostic Dump against the step-by-step workflow, or explicitly label them as quick-reference recap to justify the repetition.

Extract the self-contained 'Fix: Local Host Inference Timeouts (Firewall)' section into a references/ file (e.g. FIREWALL_TIMEOUT_FIX.md) and link to it from the main workflow, reducing the main body length.

Tighten or remove the 'Why CoreDNS Is Not the Cause' paragraph unless it directly resolves a commonly-misapplied fix, since it explains a non-issue rather than an action.

Dimension	Reasoning	Score
Conciseness	The body assumes Claude's competence and sticks to operational specifics, but the Common Failure Patterns table largely restates the workflow's per-step interpretation and the Full Diagnostic Dump re-lists commands already shown, adding redundancy that could be tightened beyond a minor trim.	3 / 5
Actionability	Provides copy-paste-ready openshell commands at every step, concrete provider create/update fix examples, and explicit interpretation of each diagnostic output, covering the common failure cases fully.	5 / 5
Workflow Clarity	Steps 0-7 are clearly sequenced with an explicit stop-and-report checkpoint, a sandbox probe validation step, and Verify-the-Fix / If-It-Still-Fails feedback loops for error recovery.	5 / 5
Progressive Disclosure	No bundle files exist and the entire ~400-line skill is inline; internal section structure is clear, but the self-contained firewall-fix subsection and the diagnostic dump could be split into reference files and no one-level-deep bundle references are signaled.	3 / 5
	Total	16 / 20 Passed

Description

90%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong, explicit description that clearly states what it does and when to use it, with comprehensive natural trigger keywords and a distinctive OpenShell-specific niche. The only mild weakness is that the action vocabulary is essentially a single verb (debug/diagnose) rather than multiple concrete actions.

Dimension	Reasoning	Score
Specificity	Names the domain clearly (three inference paths) but offers essentially one concrete action ("Debug why...is failing") rather than a list of several distinct actions, fitting anchor 3 rather than 4.	3 / 5
Completeness	Explicitly answers both what (debug failing inference.local, direct external, or system inference) and when ("Use when the user cannot reach a local model server, has provider base URL issues, sees inference verification failures, hits protocol mismatches...") with concrete trigger phrases.	5 / 5
Trigger Term Quality	Provides comprehensive trigger coverage including natural user phrases ("inference failing", "model server unreachable", "failed to verify inference endpoint") plus product names (ollama, vllm, sglang, trtllm, NIM) and synonyms, matching the comprehensive anchor.	5 / 5
Distinctiveness Conflict Risk	Occupies a clear OpenShell-specific niche with highly distinctive tokens (inference.local, sandbox-system, host.openshell.internal) that minimize conflict with other skills.	5 / 5
	Total	18 / 20 Passed

Validation

100%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: NVIDIA/OpenShell
Path: .agents/skills/debug-inference/SKILL.md
Commit: 69a05eb

Reviewed: 1 day ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.