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

64%

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

61%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.

The body is a well-organized, actionable diagnostic reference with concrete thresholds and examples, but it is moderately verbose, inlines time-sensitive metadata, and lacks explicit validation feedback loops for its destructive recovery steps. Progressive disclosure is good but leaves one bundle file unlinked and over-inlines detail.

Suggestions

Add explicit validate->fix->retry checkpoints to the poisoning/recovery workflow (e.g. after truncation, re-run the failing task on a small context to confirm the symptom clears before reloading full context).

Reference scripts/degradation_detector.py from the body with a 'Read when' clause, and consider moving the benchmark tables and Counterintuitive Findings into references/ to slim SKILL.md toward an overview.

Move the Created/Last Updated/Version metadata out of the body (or into a clearly marked, non-load-bearing footer) so stale-prone time-sensitive content does not compete with the active guidance.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient and well-structured, but it carries explanatory prose on mechanics Claude already knows and inlines time-sensitive content ("Created: 2025-12-20", "Version: 2.1.0", and stale-prone RULER stats) outside any deprecated/old-patterns section, which the rubric penalizes.	3 / 5
Actionability	Though instruction-only, it gives concrete actionable guidance — numeric thresholds (compaction at 70% of onset, >70% utilization), a Four-Bucket framework with explicit 'Use when' triggers, and structured YAML/text examples covering the common diagnostic cases.	4 / 5
Workflow Clarity	A detect-then-mitigate sequence is present and the Four-Bucket framework maps symptoms to strategies, but there are no explicit validate->fix->retry checkpoints, and the recommended destructive recovery operations (truncate to before the poisoning point, restart) lack validation steps, capping this dimension at 3.	3 / 5
Progressive Disclosure	Structure is good with clear sections and a one-level-deep, well-signaled reference (patterns.md with a 'Read when' clause), but the bundled scripts/degradation_detector.py is never referenced from the body and a fair amount of detail (Counterintuitive Findings, Gotchas, benchmarks) is inlined rather than split out.	4 / 5
	Total	14 / 20 Passed

Description

67%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.

The description is a solid, mostly explicit trigger statement that names a clear niche and several concrete failure modes, but it is somewhat jargon-heavy and blends the what/when into one sentence rather than mirroring the strongest 'Use when...' pattern. It is distinguishable with only minor overlap risk against sibling context skills.

Suggestions

Split into a what-clause and a separate 'Use when...' clause, e.g. 'Diagnoses and mitigates context degradation across five patterns. Use when an agent's output quality decays over long conversations, ignores information in the middle of context, or persists errors after correction.'

Add natural user-facing trigger phrases alongside the technical terms (e.g. 'agent forgets earlier instructions', 'quality drops in long chats') so it matches what users actually say.

Sharpen the action verbs beyond 'diagnosing and mitigating' (e.g. 'detect, localize, and recover from') to push specificity toward comprehensive.

Dimension	Reasoning	Score
Specificity	Names the domain and enumerates six concrete failure phenomena plus two actions ("diagnosing and mitigating"), giving broad coverage, though the action verbs themselves remain generic rather than many distinct concrete operations.	4 / 5
Completeness	It answers both what (diagnose/mitigate context degradation and its failure types) and when via the explicit "This skill should be used for..." trigger clause with concrete trigger phrases, though what and when are merged into a single sentence rather than cleanly separated.	4 / 5
Trigger Term Quality	It surfaces relevant domain terms ("lost-in-middle failures", "context poisoning", "context clash") but leans on technical jargon and misses the natural phrasings a user would actually say (e.g. "agent forgets things in long conversations", "ignores info in the middle").	3 / 5
Distinctiveness Conflict Risk	It carves a clear niche (degradation diagnosis/mitigation) with distinct triggers, but the broad framing of "agent performance degradation caused by accumulated or conflicting context" carries minor overlap risk with adjacent context-optimization/compression skills.	4 / 5
	Total	15 / 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: 6dbe1a1

Reviewed: 19 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.