CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

debug-inference

Debug inference clients that use an attached provider and its native endpoint, including hosted APIs and host-local Ollama, vLLM, SGLang, TRT-LLM, LM Studio, or NIM. Use for provider attachment, endpoint policy, credential substitution, topology, and migration from the removed managed inference endpoint. Trigger keywords - debug inference, managed inference endpoint, local inference, ollama, lm studio, vllm, sglang, trtllm, NIM, inference failing, model server unreachable, credential_endpoint_mismatch, host.openshell.internal.

Quality

92%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

81%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.

The body is a tightly written, command-driven diagnostic runbook: every workflow step carries executable openshell commands, validation is explicit (receipt/status waits, pending/failed reason inspection), and the failure table maps concrete error codes to fixes. The only weaknesses are minor: a few over-long table cells, a missing concrete probe example in step 4, and a dense inline failure table that could be a reference file.

Suggestions

Add a concrete minimal probe example in step 4 (e.g. a curl one-liner against a discovery endpoint via `openshell sandbox exec`) so the verification step is copy-paste ready like the rest of the workflow.

Tighten the longest failure-table cells (credential_placeholder_in_request_body, credential_placement_mismatch) to short fixes, moving detailed policy caveats to a reference file.

Give checklist item 1 ('Verify the engine from the gateway host') an explicit command or probe so 'verify' is executable rather than directive.

DimensionReasoningScore

Conciseness

The body is dense and product-specific with no padding or explanation of concepts Claude already knows, but a few passages could be tightened — e.g. the ~50-word table cell for `credential_placeholder_in_request_body` and the long `credential_placement_mismatch` fix cell — so it fits the 4 anchor ('efficient; minor instances that could be trimmed') rather than 5's 'every token earns its place'.

4 / 5

Actionability

Most steps are copy-paste executable ("openshell sandbox provider attach <sandbox> <provider> --wait --timeout 30", "openshell profile lint -f ./provider-profile.yaml", the symptom/cause/fix table), but step 4 ('Probe the exact endpoint... send a minimal inference request') and checklist item 1 ('Verify the engine from the gateway host') direct without giving a concrete probe command, which is the minor gap that distinguishes 4 from 5.

4 / 5

Workflow Clarity

The five-step diagnostic workflow is clearly sequenced with explicit validation checkpoints and error-recovery guidance — "If the result is pending, failed, withheld, or superseded, inspect its reason before launching the client" and "Launch the client after the attachment wait succeeds" — plus a host-local checklist, matching the 5 anchor's sequence + explicit validation + feedback loops + checklist.

5 / 5

Progressive Disclosure

Sections are well organized (Diagnostic Workflow, failure table, Host-Local Checklist, Reporting) with only flat external doc links and no nested references, but nothing is split out — the dense common-failures table is borderline reference material that could live in a separate file — so it matches 4's 'most content appropriately placed; minor organization gaps' rather than 5's 'content appropriately split'.

4 / 5

Total

17

/

20

Passed

Description

100%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.

The description is a strong example: third-person voice, concrete capability enumeration covering the full scope of the skill, an explicit 'Use for' clause, and a comprehensive trigger keyword list including product names, error codes, and internal hostnames. It clearly satisfies all four dimensions at the top anchor.

DimensionReasoningScore

Specificity

The description lists multiple concrete actions — "Debug inference clients", "provider attachment, endpoint policy, credential substitution, topology, and migration from the removed managed inference endpoint" — covering the domain comprehensively, matching the 5 anchor rather than 4 since no task area of the skill is left out.

5 / 5

Completeness

It explicitly answers both what ("Debug inference clients that use an attached provider and its native endpoint, including hosted APIs and host-local Ollama, vLLM, SGLang, TRT-LLM, LM Studio, or NIM") and when ("Use for provider attachment, endpoint policy, credential substitution, topology..." plus a dedicated trigger-keyword sentence), matching the 5 anchor's both-what-and-when-with-concrete-triggers pattern.

5 / 5

Trigger Term Quality

The explicit keyword list — "debug inference, managed inference endpoint, local inference, ollama, lm studio, vllm, sglang, trtllm, NIM, inference failing, model server unreachable, credential_endpoint_mismatch, host.openshell.internal" — covers natural phrasings users would say plus tool names and exact error codes, so it exceeds the 4 anchor's 'a few natural terms missing'.

5 / 5

Distinctiveness Conflict Risk

Highly distinctive triggers such as "credential_endpoint_mismatch" and "host.openshell.internal" pin the skill to a clear niche with minimal conflict risk against generic debugging or networking skills, matching the 5 anchor rather than 4's 'minor overlap risk'.

5 / 5

Total

20

/

20

Passed

Validation

100%

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

skills/debug-inference/SKILL.md

Reviewed

Table of Contents

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.