context-degradation

Language models exhibit predictable degradation patterns as context length increases. Understanding these patterns is essential for diagnosing failures and designing resilient systems.

Quality

26%

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

Fix and improve this skill with Tessl

tessl review fix ./skills/context-degradation/SKILL.md

Quality

Content

25%Scale 1-5

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

This skill reads as an educational essay about context degradation rather than an actionable skill for Claude. It extensively explains concepts (attention mechanisms, lost-in-middle phenomenon) that Claude already understands, while providing almost no concrete, executable guidance for diagnosing or mitigating degradation. The model-specific benchmarks appear fabricated (referencing non-existent models like GPT-5.2 and Gemini 3) which is actively harmful, and the entire document would benefit from being restructured into a concise diagnostic workflow with detailed references offloaded to bundle files.

Suggestions

Replace the descriptive essay format with a concrete diagnostic workflow: 'When you observe X symptom → check Y → apply Z mitigation → validate with W', including specific validation steps.

Move detailed explanations of each degradation pattern, model benchmarks, and research findings into separate bundle reference files, keeping SKILL.md as a concise overview with clear pointers.

Add executable code examples or concrete tool-call sequences for detecting and mitigating each degradation pattern, rather than abstract descriptions of strategies.

Remove or verify model-specific benchmark data (GPT-5.2, Gemini 3 Pro/Flash, Claude 4.5 appear to be fabricated model names) — inaccurate data in a skill file is a form of context poisoning, ironically the very problem this skill describes.

Dimension	Reasoning	Score
Conciseness	Extensively verbose with many explanations of concepts Claude already understands (attention mechanisms, what context poisoning is, cognitive load metaphors). The 'Core Concepts' section redundantly previews what the 'Detailed Topics' section then re-explains at length. Model benchmark tables contain specific version numbers (GPT-5.2, Gemini 3) that appear fabricated and add noise. Much of this content is conceptual explanation rather than actionable instruction.	2 / 5
Actionability	The skill is overwhelmingly descriptive rather than instructive. The two 'examples' are YAML/markdown sketches, not executable code or concrete commands. The 'Practical Guidance' section lists strategies abstractly (Write, Select, Compress, Isolate) without concrete implementation steps, specific tool calls, or copy-paste ready solutions. Guidelines are high-level advice rather than specific procedures.	2 / 5
Workflow Clarity	There is no clear multi-step workflow with sequenced actions and validation checkpoints. The guidelines section is a numbered list of general advice, not a process to follow. The 'Four-Bucket Approach' describes categories of strategies but doesn't sequence them into a workflow with decision points or validation steps. For a diagnostic/debugging skill, there's no structured triage process.	2 / 5
Progressive Disclosure	The content is a monolithic wall of text at ~400+ lines with no bundle files to offload detailed content. The model benchmark tables, detailed research findings, and extensive pattern descriptions should be in separate reference files. The 'References' section mentions internal references and related skills but no bundle files exist to support them. The Integration section references skills that aren't provided.	2 / 5
	Total	8 / 20 Passed

Description

28%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.

This description reads like an introductory sentence from a research paper rather than a functional skill description. It lacks concrete actions, natural trigger terms, and any explicit 'when to use' guidance. The domain is identifiable but the description provides almost no actionable information for Claude to determine when to select this skill over others.

Suggestions

Add a 'Use when...' clause with natural trigger phrases like 'long context problems', 'context window limits', 'lost in the middle', 'token limit', or 'prompt too long'.

Replace abstract language with concrete actions such as 'Diagnoses recall failures at different context positions, recommends chunking strategies, benchmarks retrieval accuracy across context lengths'.

Include specific file types, tools, or outputs this skill produces to distinguish it from general LLM advice or prompt engineering skills.

Dimension	Reasoning	Score
Specificity	Names the domain (language model context length degradation) but provides no concrete actions. 'Understanding these patterns' and 'diagnosing failures and designing resilient systems' are abstract goals, not specific capabilities like 'analyzes token usage' or 'benchmarks recall accuracy at different context lengths'.	2 / 5
Completeness	Has a vague 'what' (understanding degradation patterns) but no explicit 'when' clause. There is no 'Use when...' guidance, and the description reads more like an abstract for a research paper than a skill description that tells Claude when to select it.	2 / 5
Trigger Term Quality	Contains some relevant terms like 'context length', 'degradation patterns', and 'language models', but these are more technical/academic than what users would naturally say. Missing natural phrases like 'long context', 'lost in the middle', 'context window', 'token limit', or 'prompt too long'.	2 / 5
Distinctiveness Conflict Risk	The topic of context length degradation is somewhat specific, but the description is so vague in its actions that it could overlap with general LLM troubleshooting, prompt engineering, or system design skills. The niche is identifiable but not sharply delineated.	3 / 5
	Total	9 / 20 Passed

Validation

90%

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 — 10 / 11 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	10 / 11 Passed

Repository: administrakt0r/AI-Agents-Safe-Coding-Skills
Path: skills/context-degradation/SKILL.md
Commit: 57c135b

Reviewed: 3 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.