name: Context Manager
slug: context-manager
description: Manage conversation context and memory for optimal AI performance
category: meta
complexity: simple
version: "1.0.0"
author: "ID8Labs"
triggers:
- "manage context"
- "clear context"
- "summarize context"
- "context window"
- "optimize context"
tags:
- context-management
- memory
- optimization

Context Manager

The Context Manager skill helps you optimize conversation context to maintain Claude's effectiveness throughout long sessions. It tracks context window usage, identifies when to summarize or prune context, and helps you structure conversations to keep relevant information accessible while staying within token limits.

This skill is essential for complex, multi-day projects where conversation history grows large. It helps you decide what to preserve, what to summarize, and what to discard, ensuring Claude maintains awareness of important decisions and project state without hitting context limits.

Use this skill proactively during long development sessions, before starting new major features, or when you notice performance degradation due to context bloat.

Core Workflows

Workflow 1: Monitor Context Health

1. Check current context usage:
2. Token count
3. Percentage of limit
4. Recent growth rate
5. Analyze context composition:
6. How much is code?
7. How much is conversation?
8. How much is documentation?
9. Identify problematic areas:
10. Redundant information
11. Outdated references
12. Irrelevant tangents
13. Assess risk level:
14. Green: <50% usage, healthy
15. Yellow: 50-75% usage, monitor
16. Red: >75% usage, take action
17. Recommend actions if needed
18. Report status to user

Workflow 2: Summarize Long Conversation

1. Review conversation history
2. Extract key information:
3. Decisions made
4. Problems solved
5. Current project state
6. Open questions
7. Next steps
8. Organize by topic/timeline
9. Create concise summary
10. Validate with user
11. Suggest starting new thread with summary

Workflow 3: Prune Irrelevant Context

1. Identify candidates for removal:
2. Resolved issues
3. Abandoned approaches
4. Temporary debugging
5. Superseded information
6. Categorize by importance:
7. Safe to remove
8. Could summarize
9. Must preserve
10. Propose pruning plan to user
11. Execute approved removals
12. Preserve critical context
13. Verify coherence after pruning

Workflow 4: Optimize Context Structure

1. Analyze current context organization
2. Identify inefficiencies:
3. Information scattered across conversation
4. Redundant explanations
5. Lack of structure
6. Restructure for efficiency:
7. Group related information
8. Create reference sections
9. Use concise formats
10. Suggest external documentation for:
11. Architecture decisions
12. API specifications
13. Configuration details
14. Link to external docs instead of inlining
15. Validate improved efficiency

Quick Reference

Best Practices

• Monitor Proactively: Don't wait for performance issues
• Check context before starting major features
• Monitor after long debugging sessions

• Review weekly on long-running projects

• Summarize Regularly: Compress history at natural breakpoints

• End of feature development
• After resolving major issues

• Before switching contexts (dev → deployment)

• Externalize Static Info: Move unchanging content to files

• Architecture docs
• API specifications
• Code style guides

• Reference materials

• Use Structured Formats: Make information dense and scannable

• Tables instead of prose
• Bullet points instead of paragraphs

• Code blocks instead of descriptions

• Preserve Decisions: Always keep the "why"

• Why this approach was chosen
• Why alternatives were rejected

• What constraints influenced decisions

• Discard Aggressively: Be ruthless with temporary content

• Debugging exploration
• Failed experiments
• Resolved issues

• Superseded plans

• Start Fresh Strategically: Know when to begin new conversation

• After major milestones
• When switching to unrelated work
• When context is >75% full

• When performance degrades

• Document Externally: Use files for persistent knowledge

• PIPELINE_STATUS.md for project state
• DECISIONS.md for architecture choices
• TODO.md for task lists
• README.md for onboarding

Context Optimization Strategies

Strategy 1: Hierarchical Summarization

Long conversation →
  Detailed summary (50% reduction) →
    Executive summary (80% reduction) →
      Key decisions (95% reduction)

Strategy 2: Time-Based Windowing

Keep in context:
- Last 1 hour: Full detail
- Last 4 hours: Summarized
- Last day: Key decisions only
- Older: Link to external docs

Strategy 3: Topic-Based Partitioning

Separate threads for:
- Feature development
- Bug investigation
- Deployment/ops
- Architecture discussion

Link between threads as needed

Strategy 4: Progressive Disclosure

Start with:
- Current task context only

Add on demand:
- Related decisions
- Relevant code
- Background information

Remove when done

Context Health Checklist

Before starting a major task, verify:

• [ ] Context usage < 75%
• [ ] Recent decisions documented
• [ ] Obsolete information removed
• [ ] Current project state clear
• [ ] Next steps identified
• [ ] Relevant files/docs linked
• [ ] Debugging traces cleaned up

Warning Signs of Context Issues

Watch for these indicators:

• Responses get slower: Processing large context
• Information ignored: AI misses recent context
• Repetition: AI re-explains known information
• Loss of coherence: AI forgets earlier decisions
• Token limit warnings: Approaching hard limits
• Degraded accuracy: Mistakes in previously solid areas

External Memory Strategies

Move these to files, not context:

Context Templates

Project State Summary Template

## Project: [Name]
- **Status**: [Current pipeline stage]
- **Current focus**: [What we're working on]
- **Last completed**: [Recent achievement]
- **Next steps**: [Immediate tasks]
- **Blockers**: [What's preventing progress]
- **Key decisions**: [Recent important choices]

Decision Log Template

## Decision: [Topic]
- **Date**: [When]
- **Context**: [Why we needed to decide]
- **Options considered**: [Alternatives]
- **Choice**: [What we decided]
- **Rationale**: [Why this choice]
- **Consequences**: [Trade-offs accepted]

Session Summary Template

## Session Summary: [Date]
- **Duration**: [How long]
- **Accomplished**: [What we built/fixed]
- **Decisions**: [Choices made]
- **Issues found**: [Problems discovered]
- **Next session**: [Where to continue]

Advanced: Context Compression Techniques

For power users:

1. Use references: Link to code instead of pasting
2. "See function processData in /src/utils/data.ts"

3. Instead of: [pasting entire function]

4. Leverage AI memory: Store in knowledge graph

5. Key relationships between entities
6. Project-specific terminology

7. Team member roles and expertise

8. Create abbreviations: Define once, use everywhere

9. "FE" = Frontend, "BE" = Backend
10. "MR" = Merge Request, "PR" = Pull Request

11. Project-specific acronyms

12. Use diff format: Show changes, not entire files

13. Especially for code reviews

14. Before/after comparisons

15. Batch similar information: Group related items

16. All env vars in one block
17. All API endpoints in table
18. All dependencies in list