miles-rl-training

Provides guidance for enterprise-grade RL training using miles, a production-ready fork of slime. Use when training large MoE models with FP8/INT4, needing train-inference alignment, or requiring speculative RL for maximum throughput.

Quality

71%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/ml-inference/miles/SKILL.md

Quality

Content

63%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.

A content-rich, actionable SKILL body with concrete commands and decently sequenced workflows, undermined by significant duplication across sections and a failure to use the provided references/ bundle files. Linking the inline configuration and troubleshooting material to the existing reference files would sharply improve both conciseness and progressive disclosure.

Suggestions

Remove the duplicated speculative-decoding flag blocks and merge the two 'Key Features' sections into one; reference references/api-reference.md once instead of repeating the configuration tables inline.

Link references/troubleshooting.md from the 'Common Issues and Solutions' section instead of inlining overlapping symptom/solution pairs, and add a verification checklist to Workflow 2 to match Workflow 1.

Reconcile flag usage: surface --use-r3 and --use-tis in the main workflows where R3/TIS are described, or explicitly note they belong in the troubleshooting reference.

Dimension	Reasoning	Score
Conciseness	Noticeably redundant: speculative-decoding flags appear three times and 'Key Features' (lines 41-56) heavily overlaps 'Key Features (Conceptual)' (lines 215-252); memory and supported-models tables are also repeated. No beginner-concept padding, which keeps it above a 2.	3 / 5
Actionability	Provides many concrete, copy-paste-ready bash commands for install, quick start, and both workflows. Minor gaps: --use-r3 appears only in troubleshooting, and the 'Conceptual' section hedges that 'specific CLI flags may vary'.	4 / 5
Workflow Clarity	Both workflows have prerequisites checklists and numbered steps, and Workflow 1 includes a verification checklist; Workflow 2 lacks an explicit validation checkpoint, and the Common Issues section provides symptom-to-solution feedback loops.	4 / 5
Progressive Disclosure	Section headers give some structure, but the existing references/ files (api-reference.md, troubleshooting.md) are never linked from the body, and large config/feature/troubleshooting blocks are inlined and duplicated rather than split out; the only reference points cross-skill to ../slime/.	3 / 5
	Total	14 / 20 Passed

Description

78%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, specific description with an explicit 'Use when' clause and concrete trigger terms, clearly distinguishing miles from upstream slime and alternatives. The only soft spot is that the 'what' is framed as 'provides guidance' rather than crisp action verbs.

Dimension	Reasoning	Score
Specificity	Names the domain and several concrete capabilities (MoE with FP8/INT4, train-inference alignment, speculative RL) but the 'what' is framed as 'provides guidance' rather than enumerating concrete training actions.	4 / 5
Completeness	Has an explicit 'Use when...' clause with concrete triggers and a stated 'what', but the 'what' ('provides guidance for enterprise-grade RL training') is slightly soft versus listing concrete actions.	4 / 5
Trigger Term Quality	Includes natural practitioner phrases ('training large MoE models', 'FP8/INT4', 'train-inference alignment', 'speculative RL'); missing some common synonyms like RLHF, post-training, or quantization-aware training.	4 / 5
Distinctiveness Conflict Risk	Occupies a distinct niche (enterprise MoE RL fork of slime) with highly specific triggers (FP8/INT4, speculative RL) unlikely to collide with general skills.	5 / 5
	Total	17 / 20 Passed

Validation

87%

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 — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
relative_links	Relative link issues: 2 suspicious	Warning

	Total	14 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/ml-inference/miles/SKILL.md
Commit: 3a6c3a9

Reviewed: 1 day 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.