CleanExpo/.claude/skills/ctx

Context Health Check Skill

Name: ctx Triggers: /ctx (manual invocation only — not natural language) Version: 1.0.0

---

Purpose

One-line context health check. Runs the built-in /context command, summarises the result, and recommends ONE action: disconnect an idle MCP server, run /compact, run /clear, or "context clean" (no action needed). Designed to be cheap enough to run between tasks without itself adding measurable context.

This is the diagnostic counterpart to /next and /resume. Run it when the session feels slow or when you want to know whether you can keep going without compacting.

---

Commands

/ctx

Runs /context (the built-in Claude Code command), parses the output for:

• MCP overhead — total tokens used by idle MCP servers
• Conversation size — current % of the context window in use
• Free space — tokens remaining before auto-compact triggers
• Top consumer — single biggest token sink (conversation history, MCP, file reads, skills, etc.)

Then prints ONE LINE with:

• Status: 🟢 healthy / 🟡 watch / 🔴 act now
• Top consumer with token count
• ONE recommendation

Example outputs:

🟢 35% — top: conversation 42K — context clean, keep going
🟡 68% — top: MCP gmail 18K idle — disconnect gmail server
🔴 89% — top: conversation 180K — /next then /clear
🟡 71% — top: file reads 35K — /compact preserve "file paths, decisions"

---

Process Rules

1. One line output. Hard cap. No additional explanation, no breakdown, no "here are some other things you could do". One line.
2. One recommendation. Always. Even at 🟢 healthy, name the next thing to watch.
3. Disable model invocation. Never fire on natural language. Only explicit /ctx.
4. No file reads. This skill exists to MEASURE context, not consume more of it. The only tool call should be /context itself.
5. Honest thresholds:
   • 🟢 < 60% used
   • 🟡 60-79% used
   • 🔴 ≥ 80% used (auto-compact zone)

---

Recommendation Decision Tree

IF MCP servers > 5 tokens/message AND idle > 10 min:
  → "disconnect <server>" (CLI > MCP)
ELIF conversation > 70% AND task complete:
  → "/next then /clear"
ELIF conversation > 70% AND task in progress:
  → "/compact preserve <key items>"
ELIF file reads > 30K AND task complete:
  → "/compact preserve file paths"
ELIF skills loaded > 10 AND most unused:
  → "rare skills should set disable-model-invocation"
ELSE:
  → "context clean, keep going"

---

When NOT to use this skill

• You already know context is fine — skip the check
• You already know context is full — just /compact or /clear directly, no diagnosis needed
• You want a detailed breakdown — run /context directly instead

---

Related

• .claude/skills/context-monitor/SKILL.md — heavier auto-monitoring skill that runs every 10 messages
• .claude/skills/next/SKILL.md — task transition (the action /ctx recommends when conversation > 70%)
• .claude/skills/resume/SKILL.md — recovery after /clear