long-context

Extend context windows of transformer models using RoPE, YaRN, ALiBi, and position interpolation techniques. Use when processing long documents (32k-128k+ tokens), extending pre-trained models beyond original context limits, or implementing efficient positional encodings. Covers rotary embeddings, attention biases, interpolation methods, and extrapolation strategies for LLMs.

Quality

75%

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/llm-tools/long-context/SKILL.md

Quality

Content

65%

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 complete executable code and commands, but it is verbose and monolithic — re-explaining well-known concepts and inlining material that duplicates the provided reference files. Workflows also lack explicit validation checkpoints.

Suggestions

Cut the 'Core Concepts' theory explanations and mathematical formulations (concepts Claude already knows); keep the body as executable code plus decision guidance, and point to references/rope.md and references/extension_methods.md for theory.

Inline-link the reference files where relevant (e.g., next to the RoPE section link rope.md, next to the fine-tuning section link fine_tuning.md) instead of only listing them in See Also, so the body can shed duplicated content.

Add explicit validation checkpoints to the extension/fine-tuning workflow (e.g., evaluate perplexity on a long-context held-out set after scaling, and only proceed when it passes).

Dimension	Reasoning	Score
Conciseness	The ~525-line body re-explains concepts Claude already knows (RoPE/ALiBi mechanics, 'Encodes absolute position via rotation matrix', mathematical formulations) and duplicates RoPE/ALiBi/YaRN content across Quick Start and Core Concepts, fitting the score-2 anchor of mostly efficient with unnecessary explanation; it is not pure fluff (the code is genuinely useful) but not lean.	2 / 3
Actionability	Provides fully executable implementations (RotaryEmbedding, apply_rotary_pos_emb, ALiBi slopes/bias, LongContextAttention), copy-paste rope_scaling config dicts, concrete TrainingArguments, and exact CLI commands with flags, matching the score-3 anchor.	3 / 3
Workflow Clarity	Procedures are sequenced and a decision guide plus good/bad pitfalls exist, but the extension and fine-tuning workflows lack explicit validation checkpoints and feedback loops (e.g., no 'verify the extended model on long context before proceeding'), which per the rubric caps batch/heavy-operation workflows at 2.	2 / 3
Progressive Disclosure	Three real, well-scoped one-level-deep reference files exist (rope.md, extension_methods.md, fine_tuning.md), but the body inlines heavy content that overlaps them (full implementations, Core Concepts theory, fine-tuning/deployment guides) and only lists references in a terminal See Also rather than signaling them inline, fitting the score-2 anchor of content that should be separate being inline.	2 / 3
	Total	9 / 12 Passed

Description

85%

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 and complete, clearly stating both capabilities and explicit use-when triggers with a distinctive niche. Its main weakness is trigger-term quality, which leans on technical jargon over the natural phrasings a user would actually say.

Suggestions

Add common-natural phrasings such as 'long context', 'extend context', and 'context extension' alongside the technique names so the trigger matches how users actually phrase requests.

Consider trimming the jargon-heavy final sentence ('Covers rotary embeddings, attention biases, interpolation methods, and extrapolation strategies for LLMs.') which largely restates techniques already named.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions ('Extend context windows', 'processing long documents', 'extending pre-trained models', 'implementing efficient positional encodings') alongside named techniques (RoPE, YaRN, ALiBi, position interpolation), matching the score-3 anchor.	3 / 3
Completeness	Clearly answers 'what' (extend context windows via named techniques) and 'when' with an explicit 'Use when processing long documents...extending pre-trained models...implementing efficient positional encodings' clause, matching the score-3 anchor.	3 / 3
Trigger Term Quality	Includes some natural terms ('long documents', 'context windows') but leans heavily on technical jargon (rotary embeddings, attention biases, extrapolation strategies) and omits common phrasings like 'long context' or 'extend context', fitting the score-2 anchor of relevant keywords missing common variations.	2 / 3
Distinctiveness Conflict Risk	Occupies a clear niche (transformer long-context positional encoding techniques) with distinct, specific triggers unlikely to fire for unrelated skills, matching the score-3 anchor.	3 / 3
	Total	11 / 12 Passed

Validation

81%

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 — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
skill_md_line_count	SKILL.md is long (538 lines); consider splitting into references/ and linking	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
referenced_paths_exist	Referenced path issues: 1 missing	Warning

	Total	13 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/llm-tools/long-context/SKILL.md
Commit: 52845c3

Reviewed: about 13 hours 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.