eagle3-new-model

Add a new model to the EAGLE3 offline pipeline. Generates an hf_offline_eagle3.yaml launcher config for a new model checkpoint, choosing the right hidden state dump backend (TRT-LLM / HF / vLLM) and GPU configuration. Use when user wants to run EAGLE3 on a model that does not yet have a YAML in tools/launcher/examples/ or asks how to configure the pipeline for a new checkpoint.

Quality

93%

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

Quality

Content

87%Weight 40%Scale 1-3

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

The body is lean and highly actionable, with concrete paths, scripts, and per-model knobs in efficient tables. Its one weakness is workflow clarity: it delegates the multi-step pipeline ordering to external examples and provides only a single --dryrun checkpoint for a costly batch operation.

Suggestions

Include a brief inline sequence of the four tasks (task_0 → task_1 → task_2 → task_3) with a validation checkpoint between each, rather than fully deferring ordering to the reference example.

Add an explicit validate/feedback loop for the batch train step (e.g., check task_2 logs for convergence errors before launching task_3) to satisfy the rubric's feedback-loop requirement for destructive/batch operations.

Clarify that --dryrun should be re-run after any model-specific adjustment to confirm the full 4-task config still resolves before submission.

Dimension	Reasoning	Score
Conciseness	The ~35-line body is lean, assumes Claude's competence ('This file documents only the two things that are not obvious from the examples'), and avoids explaining concepts Claude already knows; tables and prose are token-efficient.	3 / 3
Actionability	Concrete executable guidance throughout: exact target path, a backend-selection table with real script paths, and specific knobs (--trust-remote-code, --tp <TP>, --moe-ep <EP>, TIKTOKEN_RS_CACHE_DIR, intermediate_size).	3 / 3
Workflow Clarity	The 4-task pipeline sequence (task_0..task_3) is deliberately deferred to the reference examples rather than sequenced inline, and only one soft checkpoint (--dryrun preview) is given for an expensive batch/train workflow, which the rubric caps at 2.	2 / 3
Progressive Disclosure	Under 50 lines, single-purpose, and organized into clear sections with two well-structured tables; no external bundle files are needed, which meets the rubric's simple-skill allowance for a top score.	3 / 3
	Total	11 / 12 Passed

Description

100%Weight 40%Scale 1-3

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, complete, and well-scoped: it states concrete capabilities, gives an explicit 'Use when' trigger, and carves out a distinct niche unlikely to conflict with other skills. It is concise with no padding or over-claims.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'Generates an hf_offline_eagle3.yaml launcher config', 'choosing the right hidden state dump backend (TRT-LLM / HF / vLLM) and GPU configuration' — rather than vague language.	3 / 3
Completeness	Explicitly answers both what it does (generates the YAML config, selects dump backend and GPU config) and when to use it via an explicit 'Use when' clause naming two trigger scenarios.	3 / 3
Trigger Term Quality	Natural phrasings a user would say are present: 'Use when user wants to run EAGLE3 on a model', 'asks how to configure the pipeline for a new checkpoint', covering common request variations.	3 / 3
Distinctiveness Conflict Risk	A narrow EAGLE3-offline-pipeline niche with triggers tied to a concrete path condition (no YAML in tools/launcher/examples/) makes it unlikely to fire for unrelated skills.	3 / 3
	Total	12 / 12 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
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: NVIDIA/Model-Optimizer
Path: .agents/skills/eagle3-new-model/SKILL.md
Commit: 33d05b0

Reviewed: 15 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.