Tessl • The package manager for agent skills

context-degradation

This skill should be used for diagnosing and mitigating context degradation: lost-in-middle failures, context poisoning, context clash, context confusion, attention-pattern issues, and agent performance degradation caused by accumulated or conflicting context.

Quality

73%

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 ./skills/context-degradation/SKILL.md

Quality

Content

57%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 well-researched, genuinely informative body with concrete thresholds and a useful symptom-keyed mitigation framework, but it is held back by systematic content duplication between Core Concepts and Detailed Topics, a diagnosis workflow that is never laid out as a sequenced procedure with validation checkpoints, and progressive-disclosure gaps (inline detail that belongs in the reference, plus an unreferenced script).

Suggestions

Merge each pattern's Core Concepts paragraph with its Detailed Topics section — every pattern (poisoning, distraction, confusion, clash) is currently explained twice, and the duplicated halves can be combined into one section each, cutting the body by roughly a third.

Add a short sequenced 'Diagnosis procedure' section (e.g., 1. rule out prompt quality by testing at low context length, 2. establish a baseline over multiple runs, 3. match symptoms to a pattern, 4. apply the matching bucket's mitigation) so the workflow's validation checkpoints are explicit steps rather than buried in Gotchas 1 and 5.

Reference scripts/degradation_detector.py from the body (it is bundled but currently orphaned), and move the Empirical Benchmarks / Counterintuitive Findings detail into references/patterns.md, keeping only the headline thresholds inline.

Dimension	Reasoning	Score
Conciseness	Each of the five patterns is explained twice — once in Core Concepts and again in Detailed Topics (e.g., poisoning at "Treat context poisoning as a circuit breaker problem" and again under "Context Poisoning: Prevention and Recovery"), and the 8-item Guidelines section restates points already made. The material itself is non-obvious and research-backed, but the systematic duplication means it could be tightened by roughly a third.	3 / 5
Actionability	Provides concrete, usable guidance: numeric thresholds ("exceeds 70% of the window", "degradation begins at 8K-16K tokens", compaction "at 70% of known onset"), a four-bucket framework with per-pattern use-when triggers, and structured examples like the clash annotation and poisoning circuit-breaker steps. For an instruction-only skill this is mostly executable; it falls short of 5 because examples like Example 1 are illustrative rather than directly usable.	4 / 5
Workflow Clarity	The diagnose-to-mitigate flow exists (detection signals per pattern, then the four-bucket mitigation framework keyed to symptoms) but is scattered across sections rather than presented as a sequenced procedure. Validation checkpoints are mostly implicit — the one genuine pre-diagnosis check ("verify the same prompt works correctly at low context lengths") is buried in Gotcha 5 rather than placed as an explicit step in the workflow.	3 / 5
Progressive Disclosure	The one-level reference to ./references/patterns.md is well signaled with a "Read when" trigger, but the body inlines ~40 lines of benchmark and counterintuitive-findings detail that belongs in that reference, and the bundled scripts/degradation_detector.py is never referenced from SKILL.md, leaving a bundle file orphaned from navigation.	3 / 5
	Total	13 / 20 Passed

Description

75%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 description that clearly scopes the domain, enumerates concrete failure patterns, and uses an explicit activation clause with natural trigger vocabulary. Its main limitations are a when-clause that duplicates the what rather than adding distinct trigger situations, and no disambiguation from closely related skills in the same collection.

Dimension	Reasoning	Score
Specificity	Names concrete actions ("diagnosing and mitigating") and enumerates five specific failure patterns ("lost-in-middle failures, context poisoning, context clash, context confusion, attention-pattern issues"), giving comprehensive coverage of the domain. Not a 5 because the actions themselves are generic verbs rather than multiple distinct concrete capabilities.	4 / 5
Completeness	Answers both what ("diagnosing and mitigating context degradation" plus the pattern list) and when via the explicit "This skill should be used for..." clause. Not a 5 because the when-clause merely restates the what instead of naming concrete user situations or triggers that would fire the skill.	4 / 5
Trigger Term Quality	Includes natural phrases users would actually say — "lost-in-middle failures", "context poisoning", "agent performance degradation" — which map well to real troubleshooting language. A few natural variations are missing (e.g., "long context failures", "context rot", "degrading over a long conversation").	4 / 5
Distinctiveness Conflict Risk	The niche (context failure diagnosis/mitigation) is clear and the pattern names are distinctive triggers. Minor overlap risk remains with adjacent skills in the same collection (context-fundamentals, context-optimization, context-compression) that own nearby territory, which the description does not disambiguate.	4 / 5
	Total	16 / 20 Passed

Validation

100%

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 — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: muratcankoylan/Agent-Skills-for-Context-Engineering
Path: skills/context-degradation/SKILL.md
Commit: 58b55a8

Reviewed: 11 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.