Context Health Check

Give the user an instant, single-line status of their context window. No tables, no analysis, no interruption.

Output format

Produce exactly one status line followed by one action line:

{emoji} {X}% full (~{used} / {total} tokens) — {status message}
💡 {one specific next step}

Emoji and status message rules

Bloat detection addendum

After the fill level message, append one of:

• — no bloat detected if everything in context looks relevant to the current task
• — {N} bloat item(s) detected if you spot files, history, or instructions clearly unrelated to what the user is doing right now

Action line rules

Pick the most relevant single action:

• If < 40% full and no bloat: You're good — no action needed.
• If bloat detected: Run /token-inspector:find-bloat to identify what to remove.
• If > 65% full: Run /token-inspector:inspect-context for a full breakdown.
• If > 85% full: Run /token-inspector:find-bloat now or start a new conversation to avoid truncation.
• If > 95% full: Start a new conversation immediately — responses may be truncated.

Estimation

• Estimate total tokens used by summing: system prompt, CLAUDE.md, loaded files, conversation history, skills, MCP tool definitions, embedded documents.
• Use 1 token ≈ 4 characters for prose, slightly fewer for dense code.
• For context window size: assume 200,000 tokens unless you have evidence of a different limit.

Examples of valid output

🟢 18% full (~36,000 / 200,000 tokens) — Context is healthy — no bloat detected
💡 You're good — no action needed.

🟠 71% full (~142,000 / 200,000 tokens) — Context is getting crowded — 3 bloat item(s) detected
💡 Run /token-inspector:find-bloat to identify what to remove.

🚨 94% full (~188,000 / 200,000 tokens) — Context is critical — 5 bloat item(s) detected
💡 Start a new conversation immediately — responses may be truncated.

Keep it to two lines. Do not add headers, tables, bullet lists, or extra explanation.

$ARGUMENTS