/context-health

Analyze the current session context for health issues including degradation, poisoning indicators, and optimization opportunities.

System Prompt

EXECUTION DIRECTIVE: When the user invokes this command, you MUST:

1. IMMEDIATELY analyze the current session context
2. Calculate health metrics (token usage, degradation risk, poisoning indicators)
3. Provide recommendations for improvement

DO NOT:

• Skip analysis steps
• Provide generic responses without actual metrics
• Modify any context (read-only analysis)

---

Usage

/context-health [options]

Options

Option	Type	Default	Description
--verbose	boolean	false	Show detailed metrics breakdown
--recommendations	boolean	true	Include improvement suggestions
--format	string	text	Output format (text, json, markdown)

Analysis Components

1. Token Utilization

• Current token count (estimated)
• Percentage of context limit used
• Category breakdown (system prompt, conversation, tool outputs)

2. Attention Degradation

• Identify "lost-in-middle" risk areas
• Map critical information positions
• Flag attention-degraded regions (middle 60% of context)

3. Context Poisoning

• Error accumulation check
• Contradiction detection
• Hallucination marker identification

4. Health Score

• Composite score (0.0 - 1.0)
• Status classification (Healthy/Warning/Degraded/Critical)

Output Format

Standard Output

CONTEXT HEALTH REPORT
=====================
Health Score: 0.82
Status: Healthy

Metrics:
├── Token Utilization: 45%
├── Degradation Risk: Low
└── Poisoning Risk: None

Recommendations:
1. Context is healthy - continue monitoring

Verbose Output (--verbose)

CONTEXT HEALTH REPORT (DETAILED)
================================
Health Score: 0.72
Status: Warning

Token Analysis:
├── Total Tokens: ~36,000 / 80,000
├── System Prompt: 2,500 (7%)
├── Conversation: 18,000 (50%)
├── Tool Outputs: 15,500 (43%)
└── Utilization: 45%

Attention Distribution:
├── Beginning (0-10%): 8,000 tokens [HIGH ATTENTION]
├── Middle (10-90%): 64,000 tokens [DEGRADED ATTENTION]
└── End (90-100%): 8,000 tokens [HIGH ATTENTION]

Critical Information Positions:
├── System prompt: Position 0-2,500 ✓ Safe
├── Task context: Position 25,000-30,000 ⚠ At Risk
└── Recent outputs: Position 72,000-80,000 ✓ Safe

Poisoning Indicators:
├── Error accumulation: 2 errors detected (Low risk)
├── Contradictions: None detected
└── Hallucination markers: None detected

Recommendations:
1. Move critical task context closer to beginning or end
2. Consider summarizing tool outputs (43% of context)
3. Monitor error accumulation as session continues

JSON Output (--format json)

{
  "health_score": 0.72,
  "status": "warning",
  "metrics": {
    "token_count": 36000,
    "utilization": 0.45,
    "degradation_score": 0.28,
    "poisoning_risk": "low"
  },
  "issues": {
    "lost_in_middle": {
      "at_risk": [25000, 30000],
      "safe": [0, 2500, 72000, 80000]
    },
    "poisoning": {
      "error_count": 2,
      "contradictions": 0,
      "hallucination_markers": 0
    }
  },
  "recommendations": [
    "Move critical task context closer to beginning or end",
    "Consider summarizing tool outputs"
  ]
}

Examples

Basic Health Check

/context-health

Quick overview with health score and status.

Detailed Analysis

/context-health --verbose

Full breakdown with position analysis and category metrics.

Export for Logging

/context-health --format json

Machine-readable output for automated monitoring.

Recommendations Only

/context-health --recommendations --no-verbose

Focus on actionable improvements.

Related Components

Invokes Agent

• context-health-analyst: Performs the actual analysis

Related Commands

• /cx: Session management (can incorporate health checks)
• /cxq: Context query (complements health analysis)

Related Skills

• context-degradation: Detection algorithms
• context-fundamentals: Core concepts
• context-optimization: Improvement strategies

Implementation Notes

This command invokes the context-health-analyst agent to perform analysis. The agent:

1. Estimates token count using ~4 chars/token heuristic
2. Simulates attention distribution based on U-shaped attention curve research
3. Scans for error, contradiction, and hallucination patterns
4. Calculates composite health score
5. Generates context-aware recommendations

When to Use

Use this command when:

• Session feels "sluggish" or responses seem less coherent
• After many tool calls with verbose outputs
• Before starting a complex multi-step task
• To validate context quality after major changes

Not needed when:

• Short, simple conversations
• Starting a fresh session
• Context is clearly minimal

Success Output

When context health check completes:

✅ COMMAND COMPLETE: /context-health
Health Score: X.XX
Status: <Healthy|Warning|Degraded|Critical>
Token Utilization: X%
Degradation Risk: <Low|Medium|High>
Recommendations: N provided

Completion Checklist

Before marking complete:

• Token count estimated
• Degradation analyzed
• Poisoning checked
• Health score calculated
• Recommendations generated

Failure Indicators

This command has FAILED if:

• ❌ No health score calculated
• ❌ No status determined
• ❌ Analysis incomplete
• ❌ No output displayed

When NOT to Use

Do NOT use when:

• Fresh session just started
• Short simple conversation
• Context clearly minimal

Anti-Patterns (Avoid)

Anti-Pattern	Problem	Solution
Ignore warnings	Context degrades	Act on recommendations
Skip after long session	Missed issues	Check periodically
No action on critical	Session failure	Summarize or restart

Principles

This command embodies:

• #9 Based on Facts - Metric-based analysis
• #6 Clear, Understandable - Clear status
• #3 Complete Execution - Full health check

Full Standard: CODITECT-STANDARD-AUTOMATION.md