context-degradation
Use when diagnosing agent failures, debugging lost-in-middle issues, understanding context poisoning, or asking about "context degradation", "lost in middle", "context poisoning", "attention patterns", "context clash", "agent performance drops"
What this skill does
# Context Degradation Patterns Language models exhibit predictable degradation as context grows. Understanding these patterns is essential for diagnosing failures and designing resilient systems. ## Degradation Patterns | Pattern | Cause | Symptoms | |---------|-------|----------| | Lost-in-Middle | Attention mechanics | 10-40% lower recall for middle content | | Context Poisoning | Errors compound | Tool misalignment, persistent hallucinations | | Context Distraction | Irrelevant info | Uses wrong information for decisions | | Context Confusion | Mixed tasks | Responses address wrong aspects | | Context Clash | Conflicting info | Contradictory guidance derails reasoning | ## Lost-in-Middle Information at beginning and end receives reliable attention. Middle content suffers dramatically reduced recall. **Mitigation**: ```markdown [CURRENT TASK] # At start (high attention) - Goal: Generate quarterly report - Deadline: End of week [DETAILED CONTEXT] # Middle (less attention) - 50 pages of data - Supporting evidence [KEY FINDINGS] # At end (high attention) - Revenue up 15% - Growth in Region A ``` ## Context Poisoning Once errors enter context, they compound through repeated reference. **Entry pathways**: 1. Tool outputs with errors 2. Retrieved docs with incorrect info 3. Model-generated summaries with hallucinations **Symptoms**: - Tool calls with wrong parameters - Strategies that take effort to undo - Hallucinations that persist despite correction **Recovery**: - Truncate to before poisoning point - Explicitly note poisoning and re-evaluate - Restart with clean context ## Context Distraction Even a single irrelevant document reduces performance. Models must attend to everything—they cannot "skip" irrelevant content. **Mitigation**: - Filter for relevance before loading - Use namespacing for organization - Access via tools instead of context ## Degradation Thresholds | Model | Degradation Onset | Severe Degradation | |-------|-------------------|-------------------| | GPT-5.2 | ~64K tokens | ~200K tokens | | Claude Opus 4.5 | ~100K tokens | ~180K tokens | | Claude Sonnet 4.5 | ~80K tokens | ~150K tokens | | Gemini 3 Pro | ~500K tokens | ~800K tokens | ## The Four-Bucket Approach | Strategy | Purpose | |----------|---------| | **Write** | Save context outside window | | **Select** | Pull relevant context in | | **Compress** | Reduce tokens, preserve info | | **Isolate** | Split across sub-agents | ## Counterintuitive Findings 1. **Shuffled haystacks outperform coherent** - Coherent context creates false associations 2. **Single distractors have outsized impact** - Step function, not proportional 3. **Needle-question similarity matters** - Dissimilar content degrades faster ## When Larger Contexts Hurt - Performance degrades non-linearly after threshold - Cost grows exponentially with context length - Cognitive bottleneck remains regardless of size ## Best Practices 1. Monitor context length and performance correlation 2. Place critical information at beginning or end 3. Implement compaction triggers before degradation 4. Validate retrieved documents for accuracy 5. Use versioning to prevent outdated info clash 6. Segment tasks to prevent confusion 7. Design for graceful degradation 8. Test with progressively larger contexts
Related in AI Agents
skill-development
IncludedComprehensive meta-skill for creating, managing, validating, auditing, and distributing Claude Code skills and slash commands (unified in v2.1.3+). Provides skill templates, creation workflows, validation patterns, audit checklists, naming conventions, YAML frontmatter guidance, progressive disclosure examples, and best practices lookup. Use when creating new skills, validating existing skills, auditing skill quality, understanding skill architecture, needing skill templates, learning about YAML frontmatter requirements, progressive disclosure patterns, tool restrictions (allowed-tools), skill composition, skill naming conventions, troubleshooting skill activation issues, creating custom slash commands, configuring command frontmatter, using command arguments ($ARGUMENTS, $1, $2), bash execution in commands, file references in commands, command namespacing, plugin commands, MCP slash commands, Skill tool configuration, or deciding between skills vs slash commands. Delegates to docs-management skill for official documentation.
reprompter
IncludedTransform messy prompts into well-structured, effective prompts — single or multi-agent. Use when: "reprompt", "reprompt this", "clean up this prompt", "structure my prompt", rough text needing XML tags and best practices, "reprompter teams", "repromptception", "run with quality", "smart run", "smart agents", multi-agent tasks, audits, parallel work, anything going to agent teams. Don't use when: simple Q&A, pure chat, immediate execution-only tasks. See "Don't Use When" section for details. Outputs: Structured XML/Markdown prompt, quality score (before/after), optional team brief + per-agent sub-prompts, agent team output files. Success criteria: Single mode quality score ≥ 7/10; Repromptception per-agent prompt quality score 8+/10; all required sections present, actionable and specific.
adaptive-compaction
IncludedAdaptive add-on policy and recovery layer that decides WHEN to compact, prune, snapshot, or fork -- replacing fixed-percent auto-compaction across Claude Code, Codex, and MCP-capable hosts. Trigger on auto-compact timing or damage: "when should I compact", "is it safe to compact now or start a fresh session", "auto-compact fires too early/mid-task", "switching to an unrelated task but the window still has space", "context rot", "answers get worse the longer the session runs", "the agent forgot the plan or my decisions after it summarized", "add a layer on top that manages context without changing the agent", raising autoCompactWindow to give the policy room, or installing/tuning a cross-tool compaction policy or PreCompact hook -- even when "compaction" is never said but the problem is context-window pressure or post-summarization memory loss. Do NOT use to summarize a conversation, build RAG, write a summarization prompt (decides WHEN not HOW), or answer max-context-length trivia.
agent-skill-creator
IncludedCreate cross-platform agent skills from workflow descriptions. Activates when users ask to create an agent, automate a repetitive workflow, create a custom skill, or need advanced agent creation. Triggers on phrases like create agent for, automate workflow, create skill for, every day I have to, daily I need to, turn process into agent, need to automate, create a cross-platform skill, validate this skill, export this skill, migrate this skill. Supports single skills, multi-agent suites, transcript processing, template-based creation, interactive configuration, cross-platform export, and spec validation.
llm-wiki
IncludedUse when building or maintaining a persistent personal knowledge base (second brain) in Obsidian where an LLM incrementally ingests sources, updates entity/concept pages, maintains cross-references, and keeps a synthesis current. Triggers include "second brain", "Obsidian wiki", "personal knowledge management", "ingest this paper/article/book", "build a research wiki", "compound knowledge", "Memex", or whenever the user wants knowledge to accumulate across sessions instead of being re-derived by RAG on every query.
skill-master
IncludedAgent Skills authoring, evaluation, and optimization. Create, edit, validate, benchmark, and improve skills following the agentskills.io specification. Use when designing SKILL.md files, structuring skill folders (references, scripts, assets), ingesting external documentation into skills, running trigger evals, benchmarking skill quality, optimizing descriptions, or performing blind A/B comparisons. Keywords: agentskills.io, SKILL.md, skill authoring, eval, benchmark, trigger optimization.