CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

offline-packing-env-vars

Bilingual guide for the OFFLINE_PACKING_BMR and OFFLINE_PACKED_DATA environment variables that control LLaVA-OneVision2 training-side packing — what each gate does, why both must be enabled together, MBS=1 requirement, and the dead OFFLINE_PACKING_VQA branch

Quality

77%

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

Fix and improve this skill with Tessl

tessl review fix ./.opencode/skills/offline-packing-env-vars/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

75%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 highly actionable with exact code and line references, clear gate logic, and a strong diagnostics checklist. Its main weakness is token efficiency: bilingual duplication of every section roughly doubles length without adding new information.

Suggestions

Collapse the bilingual duplication: keep one primary language inline and move the other to a bundled reference file (e.g. references/zh.md) to reclaim roughly half the tokens.

Consider extracting the full BMR×PACKED_DATA truth table and end-to-end flow diagram into a references/ file, keeping SKILL.md as a lean overview that points to it.

Add an explicit validate-then-proceed loop to the Recipe section (e.g. 'after exporting, run the diagnostic prints and confirm [1,P+1] before launching training') to strengthen the primary workflow's checkpoints.

DimensionReasoningScore

Conciseness

The English prose is dense, code-heavy, and assumes Claude's competence with no basic-concept padding, but the full bilingual duplication of every section roughly doubles the token load, which is noticeable rather than minor overhead.

3 / 5

Actionability

Provides copy-paste-ready bash exports, exact file:line references, verbatim Python snippets from the consumer side, and concrete diagnostics (grep commands, exact print statements at named lines) covering the common cases.

5 / 5

Workflow Clarity

Clear end-to-end flow diagram, an exhaustive BMR×PACKED_DATA truth table, and a numbered diagnostics checklist with explicit validation steps (print cu_lengths.shape, expect [1,P+1]); minor gap is that the primary recipe is simple rather than a multi-step validated workflow.

4 / 5

Progressive Disclosure

Well-organized with clear section headers, a source-file index, and one-level cross-references to companion skills; no bundle files exist so all content is inline in one cohesive file, with minor room to split reference material out.

4 / 5

Total

16

/

20

Passed

Description

70%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 specific, niche-scoped, and rich in concrete trigger terms, but it omits an explicit 'Use when...' trigger clause, capping completeness. Adding a trigger phrase would raise the completeness score.

Suggestions

Append an explicit 'Use when...' clause, e.g. 'Use when configuring or debugging LLaVA-OneVision2 training shell scripts that enable sample packing.'

Lead with the action framing ('Explains and diagnoses...') before the topic list so the 'what Claude does' is as concrete as the 'what it covers'.

Consider adding natural synonyms like 'packing env vars' or 'padding-free attention' alongside the exact variable names to broaden trigger coverage.

DimensionReasoningScore

Specificity

Lists several concrete coverage areas — 'what each gate does', 'why both must be enabled together', 'MBS=1 requirement', and 'the dead OFFLINE_PACKING_VQA branch' — giving comprehensive topical scope, though framed as topics a guide covers rather than discrete actions Claude performs.

4 / 5

Completeness

Clearly states what the guide covers, but there is no 'Use when...' clause or equivalent explicit trigger guidance, which caps completeness at 3 per the rubric guidelines.

3 / 5

Trigger Term Quality

Names the exact env vars ('OFFLINE_PACKING_BMR', 'OFFLINE_PACKED_DATA') and 'training-side packing' that a user debugging this codebase would naturally grep for or say, with good though not exhaustive keyword coverage.

4 / 5

Distinctiveness Conflict Risk

Targets a highly specific niche (two named env vars in LLaVA-OneVision2 training-side packing) with distinct triggers and minimal overlap risk with other skills.

5 / 5

Total

16

/

20

Passed

Validation

93%

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

CriteriaDescriptionResult

metadata_version

'metadata.version' is missing

Warning

Total

15

/

16

Passed

Repository

EvolvingLMMs-Lab/LLaVA-OneVision-2

Path

.opencode/skills/offline-packing-env-vars/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.