CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

veomni-seed-omni

SeedOmni development guidance for Open-VeOmni. Use when adding or modifying `veomni/models/seed_omni/`, Omni modules, module mixins, CPU preprocessors, conversation carrier handling, training or generation graph YAML, SeedOmni split-checkpoint scripts, per-module accelerators, distributed/eager Omni inference, or SeedOmni validation. Triggers include: new OmniModule, modify SeedOmni module, wire graph, update `training_graph` or `generation_graph`, split SeedOmni checkpoint, add Preprocessor, change conversation_list flow, configure per-module FSDP2/DDP/emb/ep, or debug SeedOmni train/infer graph behavior.

Quality

89%

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

82%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.

An exemplary routing-style SKILL.md: lean, well-organized, with a clear task-to-reference table and all reference files present. The main defects are dangling bundle references (missing `templates/` and `.agents/` paths) and the absence of fallback guidance for tasks that match no routing row.

Suggestions

Fix the dangling references: four routing rows cite `templates/*.yaml` and one cites `.agents/knowledge/veomni-seed-omni.md`, but neither exists in the skill bundle — add the template files or correct the paths so navigation never dead-ends.

Add a fallback instruction for tasks that match multiple routing rows or none (e.g., 'if no row matches or several do, start with `references/source-of-truth.md`') to close the workflow-clarity gap.

State in one line at the top what the skill itself does (routes SeedOmni tasks to the references and pins hard constraints), which would also strengthen the description's generic 'development guidance' framing.

DimensionReasoningScore

Conciseness

The ~60-line body is lean with zero padding: no general-knowledge explanations, terse factual baseline bullets ("Training graphs are flat edge lists over `module[.method]` endpoints"), and every section (First Checks, Routing, Baseline, Templates, Siblings) earns its tokens. This matches anchor 5 ('assumes Claude's competence; every token earns its place') rather than anchor 4, which requires identifiable over-explanation to trim — there is none.

5 / 5

Actionability

As an instruction-only routing skill it gives concrete, executable guidance: a task→file routing table with exact paths, named live modules to use as implementation maps ("live `fake_model`, Janus, and Qwen modules"), and a validation routing row. However, four routing rows point to `templates/*.yaml` files and one to `.agents/knowledge/veomni-seed-omni.md`, none of which exist in the bundle, so following them dead-ends — a minor gap that fits anchor 4 ('mostly executable guidance... minor gaps') rather than anchor 5.

4 / 5

Workflow Clarity

The sequence is clear: confirm the task touches SeedOmni (with concrete check paths), treat live code as source of truth, pick the smallest reference set from the table, and a dedicated validation row routes risky verification. It falls short of anchor 5 because there are no explicit checkpoints for edge cases — nothing says what to do when a task matches multiple rows or none, or when to re-check after reading a reference — matching anchor 4's 'clear sequence with most checkpoints present; minor validation gaps'.

4 / 5

Progressive Disclosure

The structure itself is exemplary: a pure overview entrypoint with a well-signaled routing table and seven one-level-deep reference files, all of which exist in `references/`. Scored against the actual bundle, though, the `templates/` directory referenced by four routing rows and `.agents/knowledge/veomni-seed-omni.md` are missing, leaving dangling navigation paths — a minor organization gap that fits anchor 4 rather than anchor 5's 'easy navigation'.

4 / 5

Total

17

/

20

Passed

Description

91%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 strong, highly targeted description with explicit trigger guidance and comprehensive natural keyword coverage for its domain. The only weakness is that the 'what' is a generic 'development guidance' phrase rather than a concrete statement of what the skill provides.

DimensionReasoningScore

Specificity

Lists many concrete capabilities and artifacts — "adding or modifying `veomni/models/seed_omni/`", "module mixins, CPU preprocessors", "SeedOmni split-checkpoint scripts", "per-module accelerators", "distributed/eager Omni inference" — with varied action verbs (add, modify, wire, update, split, configure, debug) giving comprehensive coverage of the niche. It clearly matches the anchor-5 example of multiple specific concrete actions rather than anchor 4's 'minor gaps in coverage'.

5 / 5

Completeness

The 'when' is explicit and excellent ("Use when..." plus "Triggers include:" with concrete triggers), but the 'what' reduces to the generic phrase "SeedOmni development guidance for Open-VeOmni" without stating what the guidance actually does (routing, architecture contracts, validation flow). This sits between anchor 4 ('when' could be more explicit — here it is more explicit) and anchor 5 (which requires a clearly stated 'what' as well), so the weak 'what' holds it at 4.

4 / 5

Trigger Term Quality

Triggers use the exact identifiers a developer in this codebase would naturally say: "new OmniModule", "update `training_graph` or `generation_graph`", "split SeedOmni checkpoint", "add Preprocessor", "change conversation_list flow", "configure per-module FSDP2/DDP/emb/ep", plus the source path `veomni/models/seed_omni/` and train/infer synonym variants. This matches anchor 5's comprehensive natural-term coverage including file paths; anchor 4 would require missing natural terms, and none are evident.

5 / 5

Distinctiveness Conflict Risk

The description is anchored to uniquely named artifacts (SeedOmni, Open-VeOmni, OmniModule, SeedOmni checkpoint, `conversation_list`) that no other skill would plausibly claim, matching anchor 5's 'clear niche with distinct triggers; minimal conflict risk'. Anchor 4's 'minor overlap risk with closely related skills' does not apply since even sibling VeOmni skills are explicitly scoped elsewhere.

5 / 5

Total

19

/

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

ByteDance-Seed/VeOmni

Path

.agents/skills/veomni-seed-omni/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.