CtrlK
BlogDocsLog inGet started
Tessl Logo

veomni-debug

Use this skill for ANY bug, error, crash, wrong output, loss divergence, gradient explosion, test failure, CUDA error, distributed training hang, checkpoint load failure, or unexpected behavior. Covers both quick fixes (clear root cause) and complex debugging (unclear cause). Trigger: 'fix bug', 'fix error', 'broken', 'crash', 'doesn't work', 'fails with', 'loss NaN', 'training hangs', 'FSDP error', 'OOM'.

72

Quality

87%

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

85%

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

A well-structured, highly actionable debugging protocol with clear phased workflows, verification gates, and codebase-specific guidance. The lone weakness is mild verbosity from coaching aphorisms that don't strictly earn their tokens.

Suggestions

Tighten conciseness by trimming coaching aphorisms ('If you can't reproduce, you don't understand it', 'Knowledge decays fast') and the quoted 'Red flags' lines down to imperative directives.

Condense the motivational framing in Phase 3 ('No random guess-and-check', 'Let me just try changing X...') into a single concise 'avoid uncontrolled experimentation' rule.

Move the standalone verification-subagent prompt out of inline prose into a clearly delimited, copy-paste block to reduce surrounding explanatory tokens.

DimensionReasoningScore

Conciseness

The body is mostly lean and codebase-specific (file paths, uv bisect commands, pitfalls, checklists), but it is padded with coaching aphorisms ('If you can't reproduce, you don't understand it', 'Knowledge decays fast') and the quoted 'Red flags' lines that could be tightened.

2 / 3

Actionability

It provides concrete commands ('git log --oneline -10', 'pytest tests/<module>/', '/veomni-review', 'make quality'), an executable uv bisect block, specific paths like 'veomni/distributed/parallel_plan.py', and a copy-paste verification-subagent prompt template.

3 / 3

Workflow Clarity

A clear Quick Path plus Full Protocol Phase 1-5 with TodoWrite tracking, explicit verification gates, regression checks, domain checklists, and feedback loops (15-min escalation, three-strike, hypothesis retry); validation is present for risky fixes so it is not capped at 2.

3 / 3

Progressive Disclosure

No bundle files exist; the single self-contained SKILL.md opens with a routing table and uses well-signaled sections with no nested-reference chains, which is appropriate organization for this debugging protocol.

3 / 3

Total

11

/

12

Passed

Description

90%

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, trigger-rich debugging description with explicit when-guidance and a clear domain niche. Its only real weakness is the second-person imperative voice, which caps the specificity score per the rubric's voice rule.

DimensionReasoningScore

Specificity

It lists many concrete problem types ('CUDA error', 'gradient explosion', 'checkpoint load failure', 'distributed training hang') plus a quick/complex distinction, which would merit 3, but the rubric mandates a one-point specificity penalty for non-third-person voice and 'Use this skill for...' is second-person imperative ('You can use this').

2 / 3

Completeness

It answers both 'what' (covers quick fixes and complex debugging across the listed failure modes) and 'when' with an explicit 'Trigger:' clause, so it is not capped at 2.

3 / 3

Trigger Term Quality

The 'Trigger:' list ('fix bug', 'broken', 'crash', 'doesn't work', 'fails with', 'loss NaN', 'FSDP error', 'OOM') covers natural phrases a user debugging would actually say, including common variations.

3 / 3

Distinctiveness Conflict Risk

Distinctive domain triggers ('FSDP error', 'OOM', 'loss NaN', 'gradient explosion', 'CUDA error', 'checkpoint load failure') carve out a clear distributed-training debugging niche unlikely to fire for unrelated skills, though the few generic terms ('broken', 'crash') are a mild overlap risk.

3 / 3

Total

11

/

12

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.

Validation16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository
ByteDance-Seed/VeOmni
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.