Context Health Skill

Monitor context window usage and recommend appropriate actions.

When to Use

• Periodically during long sessions
• Before starting complex tasks
• When user asks about context/memory
• Auto-trigger every 15-20 messages (if supported)

Execution Steps

1. Get Context Usage

   • Run /context command via Bash
   • Parse output for token usage
   • Extract key metrics:
     • Total usage (tokens and percentage)
     • Free space remaining
     • Autocompact buffer
     • Major consumers (MCP tools, messages, etc.)

2. Analyze Usage Patterns

   • Calculate message history size
   • Identify largest tool consumers
   • Check if approaching limits

3. Determine Health Status

   Healthy (<70%)

   • Status: 🟢 Healthy
   • Action: Continue normally
   • Message: "Context healthy, plenty of space"

   Caution (70-85%)

   • Status: 🟡 Caution
   • Action: Consider wrapping up current task
   • Message: "Context getting full, consider finishing current work"

   Warning (85-95%)

   • Status: 🟠 Warning
   • Action: Prepare for handoff soon
   • Message: "Context nearly full, prepare handoff after current task"

   Critical (>95%)

   • Status: 🔴 Critical
   • Action: Trigger /handoff immediately
   • Message: "Context critical, triggering handoff now"

4. Provide Recommendations

   Based on status, recommend:

   • Continue working (Healthy)
   • Finish current task, then handoff (Caution)
   • Create handoff after this message (Warning)
   • Auto-trigger handoff skill (Critical)

5. Optional Optimization Suggestions

   If context getting full:

   • "Consider disabling unused MCP servers"
   • "Large conversation history - handoff recommended"
   • "Complex task ahead - consider fresh session"

Output Format

🔍 Context Health Check

Status: [🟢/🟡/🟠/🔴] [Healthy/Caution/Warning/Critical]
Usage: [XXX]k / 200k tokens ([XX]%)
Free Space: [XX]k tokens ([XX]%)

Breakdown:
- System prompt: [X]k ([X]%)
- System tools: [X]k ([X]%)
- MCP tools: [X]k ([X]%)
- Custom agents: [X]k ([X]%)
- Messages: [X]k ([X]%)

Recommendation: [Action to take]

[Optional optimization suggestions]

Auto-Actions

If status is Critical (>95%):

• Automatically invoke /handoff skill
• Skip user confirmation
• Prepare for session restart

Integration Points

• Bash: Run /context command
• session-handoff: Auto-trigger if critical
• User notification: Alert about context state