context-degradation

Guidance for diagnosing predictable failures in language-model agents when their input becomes long or poorly arranged. It covers problems such as ignoring information in the middle, relying on wrong facts, and losing relevance.

In plain words
What is it for?
Use it when debugging long conversations, designing systems that handle large amounts of information, investigating lost-in-the-middle behavior, or comparing models by their context limits.
Why use it?
It helps identify the type and likely cause of an agent failure so you can choose a suitable fix.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/viktorbezdek/skillstack/context-degradation
Any agent
npx skills add viktorbezdek/skillstack --skill context-degradation
Clone the repo
git clone --depth 1 https://github.com/viktorbezdek/skillstack

Made for: Claude Code, Codex.

Per session 131 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,042 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00131 $0.02042
Opus 5 $0.00066 $0.01021
Sonnet 5 $0.00026 $0.00408
Haiku 4.5 $0.00013 $0.00204

Measured 2d ago against content hash 9c9af747726c, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

context-degradation scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

context-degradation/skills/context-degradation/SKILL.md · 174 lines

How it starts

The opening of the file, as written. The whole thing — 174 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Context Degradation Patterns

Language models exhibit predictable degradation patterns as context length increases. These patterns are not random failures — they follow measurable thresholds and can be systematically diagnosed and mitigated.

When to Use / Not Use

Use when:

  • Agent performance degrades unexpectedly during long conversations
  • Debugging cases where agents produce incorrect or irrelevant outputs
  • Designing systems that must handle large contexts reliably
  • Investigating "lost in middle" phenomena in agent outputs
  • Evaluating model selection based on degradation thresholds

Do NOT use when:

  • Learning context basics or theory -> use context-fundamentals
  • Compressing or summarizing context -> use context-compression
  • KV-cache optimization or context partitioning -> use context-optimization
  • Building isolated multi-agent architectures -> use multi-agent-patterns

Decision Tree

What degradation symptom are you seeing?
├── Agent ignores information from middle of context
│   └── Lost-in-Middle -> Place critical info at edges, use explicit headers
├── Agent keeps referencing wrong/incorrect facts
│   ├── Wrong facts came from tool output error -> Context Poisoning (truncate to before poison point)
│   ├── Wrong facts came from retrieved docs -> Context Poisoning (validate docs before loading)
│   └── Wrong facts came from model hallucination -> Context Poisoning (mark and re-evaluate)
├── Agent focuses on irrelevant information
│   └── Context Distraction -> Relevance filter before loading, namespacing, JIT context
├── Agent mixes requirements from different tasks
│   └── Context Confusion -> Task segmentation, clear transitions, state isolation
├── Agent receives contradictory information
│   ├── From different sources -> Context Clash (priority rules, conflict marking)
│   ├── From version conflicts -> Context Clash (version filtering)
│   └── From different perspectives -> Context Clash (explicit conflict marking)
└── Performance degrades as context grows
    └── Measure local degradation curve -> Choose mitigation from Four-Bucket Approach

Read the full file on GitHub · 174 lines

Files

What ships with it

3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 174 lines · 131 tokens per session scan A 9c9af747726c

Subscribe to this mod's changes

context-degradation is a skill published in the GitHub repository viktorbezdek/skillstack (11 stars, last pushed 2mo ago), licensed MIT. It adds 131 tokens to every session and 2,042 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

gossipcat

Use when installing or setting up gossipcat multi-agent orchestration (parallel review, consensus, adaptive dispatch) in Claude Code or Cursor. Installs the gossipcat MCP server and hands off to gossipstatus() for all live rules.

gossipcat-ai/gossipcat-ai · 50 tokens

Elephant Agent

Complete guide to what Elephant Agent is, how to use it from the CLI, how herd, wake, Personal Model, proactive curiosity, skills, and tools fit together, and how to explain Elephant Agent clearly to users, contributors, or evaluators.

agentic-in/elephant-agent · 54 tokens

libretranslate-markdown-i18n

Machine-translate Markdown / Obsidian / Quartz content into other languages using a self-hosted LibreTranslate instance, preserving frontmatter, headings, emojis, bold/italic, wikilinks (![[...]]/[[...]]) and internal links. Use whenever a user wants to auto-translate a Markdown vault, Obsidian notes, or a…

pedroiff0/awesome-skills · 127 tokens

awesome-skills-master

Master catalog orchestrator and autonomous installer for AI agents. Use when exploring, cloning, discovering, or installing skills, plugins, or MCP servers from awesome-skills without using interactive CLI tools. Guides environment detection, rule adaptation (.mdc/CLAUDE.md), MCP configuration, and strict author…

pedroiff0/awesome-skills · 66 tokens

readme-template

Standard README template for repos — professional structure with badges, overview, table of contents, features/modules, stack, installation, configuration, tests, security, structure, docs, roadmap, contribute, license, author + RepoActivity sections (Star History, repo stats) + profile GIF footer. Use when creating…

pedroiff0/awesome-skills · 72 tokens

suap-iff-api

Authenticate to and consume the SUAP IFF (Instituto Federal Fluminense) API v2 from the CLI — obtain the JWT access/refresh token via matricula+senha, then fetch student data (periodos letivos, dados do aluno, boletim). Use when the user has a suap.iff.edu.br matricula+senha and wants to extract academic data…

pedroiff0/awesome-skills · 102 tokens