conversion-issues

Investigate and fix model conversion issues in OpenVINO Frontends (ONNX, PyTorch) — triage, debugging, accuracy comparison, and pre-submission verification.

Quality

73%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./.github/agents-prototype/skills/conversion-issues/SKILL.md

Quality

Content

65%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 clean, token-efficient router skill with a sensible hub-and-spoke structure and genuinely useful ground rules (baseline-first verification, minimal fixes, mandatory tests). However, all of the referenced detail files are absent from the bundle, leaving the skill's actionable core unresolvable, and the workflow ordering exists only implicitly across sections.

Suggestions

Ship the referenced files (onnx.md, pytorch.md) in the skill bundle, or inline a minimal triage/fix workflow directly in SKILL.md so the skill is self-sufficient as delivered.

Add an explicit ordered workflow with validation checkpoints (e.g. 1. Reproduce with ONNX Runtime/PyTorch baseline → 2. Triage the failure category → 3. Compare accuracy → 4. Apply minimal fix → 5. Add test and run the full frontend suite) instead of scattering sequence hints across Notes.

Include at least one concrete, executable snippet in the body — such as the command to run the frontend test suite or build the .prototxt test model — so the hub itself carries actionable guidance even before the linked files are read.

Dimension	Reasoning	Score
Conciseness	The ~30-line body is lean and efficient: it explains nothing Claude already knows, and every table row and note carries actionable information (frontend routing, baseline verification, minimal-fix policy, test requirement). Nothing could be trimmed without losing content, matching the 'every token earns its place' anchor.	5 / 5
Actionability	Concrete guidance exists — "Read the one matching the target framework", "verify the model works with the framework's reference runtime ... before investigating OpenVINO code", "Every fix needs a test and must pass the full frontend test suite" — but the body contains no commands or code, and the files that would carry the executable workflow (onnx.md, pytorch.md, ../add-fe-op/*) are absent from the bundle. As shipped, the guidance is specific but incomplete, fitting 'some concrete guidance but incomplete; missing key details' rather than the mostly-executable anchor.	3 / 5
Workflow Clarity	Validation checkpoints are present as requirements (verify against the reference runtime first; every fix must pass the full frontend test suite), but the sequence is scattered across Goal and Notes sections rather than presented as an ordered workflow — triage → reproduce → compare → fix → test is only implicit. This matches 'steps listed but validation gaps; sequence present but checkpoints missing or implicit' rather than the clear-sequence anchor.	3 / 5
Progressive Disclosure	Judged against the actual bundle: the one-level-deep, table-signaled hub design is well-conceived (frontend workflows table plus related-skills table), but the referenced files — onnx.md, pytorch.md, and ../add-fe-op/onnx.md\|pytorch.md — do not exist in this bundle, so the core payload the overview points to is missing. The structure and signaling are clear (not buried or inlined), but the missing referenced content goes beyond the 'minor organization gaps' of the anchor above.	3 / 5
	Total	14 / 20 Passed

Description

71%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 specific, well-scoped description that clearly states what the skill does with multiple concrete actions in a distinct niche. Its main weakness is the complete absence of a 'when to use' trigger clause, which caps completeness and slightly limits natural-term coverage.

Suggestions

Add an explicit "Use when..." clause, e.g. "Use when a model fails to convert to OpenVINO IR, produces incorrect results after conversion, or when the user mentions OpenVINO, ONNX, or PyTorch conversion/translator issues."

Include natural user-facing variations such as "IR", "model export", and ".onnx" alongside the existing trigger terms.

State the boundary with the sibling add-op skill (e.g. "for implementing a new op translator from scratch, use add-fe-op instead") to reduce distinctiveness overlap risk.

Dimension	Reasoning	Score
Specificity	The description lists multiple concrete actions covering the full fix lifecycle — "Investigate and fix", "triage, debugging, accuracy comparison, and pre-submission verification" — within the named OpenVINO frontend domain. Coverage spans diagnosis through pre-submission verification with no meaningful gaps, matching the comprehensive anchor rather than the 'several actions with minor gaps' anchor below it.	5 / 5
Completeness	The 'what' is clearly and concretely answered, but there is no "Use when..." clause or any equivalent explicit trigger guidance; the 'when' is only weakly implied by the domain itself. Per the judging guidelines a missing 'Use when' clause caps completeness at 3 — it is not 2 because the 'what' is specific, not vague.	3 / 5
Trigger Term Quality	"conversion issues", "OpenVINO", "ONNX", "PyTorch", "triage", and "accuracy" are natural terms a user debugging a conversion would say. Natural variations and extensions users would plausibly use — "model fails to convert", "IR", "export", ".onnx" — are missing, which fits the 'good keyword coverage, a few natural terms missing' anchor rather than the comprehensive one.	4 / 5
Distinctiveness Conflict Risk	"model conversion issues in OpenVINO Frontends (ONNX, PyTorch)" carves out a clear, distinct niche with domain-specific triggers. Minor overlap risk remains against the closely related sibling skill for implementing new op translators (add-fe-op), whose boundary is not articulated in the description itself, so it falls just short of the 'minimal conflict risk' anchor.	4 / 5
	Total	16 / 20 Passed

Validation

93%

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 — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
relative_links	Relative link issues: 2 missing, 2 suspicious	Warning

	Total	15 / 16 Passed

Repository: openvinotoolkit/openvino
Path: .github/agents-prototype/skills/conversion-issues/SKILL.md
Commit: f766feb

Reviewed: 3 days 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.