CtrlK
BlogDocsLog inGet started
Tessl Logo

length-pool-sort-dataset

Bilingual guide for understanding LengthPoolSortDataset cross-rank length synchronization mechanism in multi-GPU training

53

Quality

60%

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/length-pool-sort-dataset/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

67%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 well-structured, actionable, and gives a clear debugging workflow with concrete code, paths, and a tuning table. Main weaknesses are token cost from full bilingual duplication and the absence of an explicit validate/retry loop, though the latter is less critical for an analytical skill.

Suggestions

Collapse the bilingual duplication by interleaving English and Chinese inline per point or moving the Chinese translation to a collapsible/reference section to halve token cost and lift conciseness.

Add a short validate/retry loop to the debugging workflow (e.g. 'After adjusting pool_size, re-measure step-time variance; if unchanged, increase pool_size 10x') to push workflow clarity to 5.

Provide a standalone, copy-paste-ready minimal reproduction snippet (concrete imports and a small runnable example) to close the minor actionability gap.

DimensionReasoningScore

Conciseness

Mostly efficient with lean code/diagrams, but full bilingual duplication roughly doubles the token count and the 'gradient all-reduce' basics are concepts Claude already knows. Not a 2 because the per-language content is tight and not padded with generic library introductions; not a 4 because the duplication and basic explanations could be trimmed.

3 / 5

Actionability

Provides concrete code (three-step pipeline, key_fn lambda), exact file paths, the CLI flag '--length-sort-pool-size N', a pipeline-position diagram, tuning tables, and a debugging checklist. Not a 5 because the pipeline snippet is illustrative (uses self.dataset/self.pool_size) rather than standalone copy-paste-ready code with minor gaps.

4 / 5

Workflow Clarity

The 'What to check during debugging' checklist and 'Expected outputs' section give a clear sequenced analysis workflow (is it active → length distribution → num_workers → step-time variance). Not a 5 because there is no explicit validate/retry feedback loop, though the skill is analytical rather than destructive so the workflow-clarity cap does not apply.

4 / 5

Progressive Disclosure

Well-organized into clearly headed sections (Purpose, Core mechanism, Why it accelerates, tuning, multi-worker, checkpoint, debugging) with no need for external references for a self-contained conceptual skill. Not a 5 because the bilingual duplication of every section and table is a minor organization gap, and no quick-start/advanced split is offered.

4 / 5

Total

15

/

20

Passed

Description

52%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 highly specific and distinctive but lacks a 'Use when...' trigger clause and states only a minimal action ('understanding'), limiting completeness and trigger guidance. Adding concrete verbs (analyze, debug, tune pool_size) and an explicit when-clause would raise both dimensions.

Suggestions

Add an explicit 'Use when...' trigger clause, e.g. 'Use when debugging multi-GPU training speed issues caused by uneven sample lengths across ranks.'

Replace the generic action 'understanding' with concrete verbs such as 'analyzes, debugs, and tunes' to lift specificity above the minimal/generic anchor.

Include natural user-facing phrases ('training is slow', 'length sort pool size', 'rank idle time') to improve trigger term quality beyond jargon-only coverage.

DimensionReasoningScore

Specificity

Names a precise domain ('LengthPoolSortDataset cross-rank length synchronization mechanism in multi-GPU training') but the only action stated is 'understanding', which is minimal and generic — matching the score-2 anchor ('Names the domain but actions are minimal or generic'). Not a 3 because no concrete actions like analyze/debug/tune are listed.

2 / 5

Completeness

Provides a clear 'what' (a bilingual guide for understanding the synchronization mechanism) but has no 'Use when...' clause or equivalent trigger guidance, capping completeness at 3 per the judging guideline. Not a 4 because the 'when' is entirely missing rather than merely implicit.

3 / 5

Trigger Term Quality

Includes relevant keywords ('multi-GPU training', 'LengthPoolSortDataset', 'cross-rank length synchronization') but they are technical jargon rather than natural phrases a user would say, and common variations/synonyms are missing. Not a 4 because natural user phrasing like 'training is slow' or 'length sort pool size' is absent.

3 / 5

Distinctiveness Conflict Risk

Targets a single named class (LengthPoolSortDataset) and a narrow mechanism (cross-rank length synchronization), forming a clear niche with distinct triggers and minimal conflict risk. Not lower because the class name is effectively unique.

5 / 5

Total

13

/

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.

Validation15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

metadata_version

'metadata.version' is missing

Warning

Total

15

/

16

Passed

Repository
EvolvingLMMs-Lab/LLaVA-OneVision-2
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.