Boost Prompt

Enhance the user's prompt using the ClaudeBoost MCP server, then present a choice.

Instructions

If `$ARGUMENTS` is empty (user just typed `/boost` with nothing after): Display this welcome message:

⚡ **ClaudeBoost** — Prompt Enhancement for Claude Code

---

💡 **Auto-boost is ON** — every task prompt you type is automatically enhanced. You don't need to type `/boost` every time — just type normally!

Example: type `fix the login bug` → ClaudeBoost auto-enhances → you choose the version you want.

---

**Commands:**
| Command | What it does |
|---------|-------------|
| Just type normally | Auto-boosts your prompt |
| `<prompt> --raw` | Skip boost for one prompt |
| `/boost <prompt>` | Manually boost a specific prompt |
| `/boost --login` | Sign in to sync history |
| `/boost --help` | Show full command reference |
| `/boost-settings` | View/change settings |
| `/boost-settings -l light` | Switch to light boost |
| `/boost-settings --auto false` | Turn off auto-boost |

STOP here. Do NOT call any MCP tools.

If `$ARGUMENTS` is `--setup` or `setup`: Get the Python path and API key, then register the MCP server:

PYTHON_PATH=$(python3 -c "from claudeboost_mcp.setup import get_python_path; print(get_python_path())")
API_KEY=$(cat ~/.claudeboost/config.env 2>/dev/null | grep ANTHROPIC_API_KEY | cut -d= -f2)

Then run:

claude mcp remove claudeboost 2>/dev/null; claude mcp add claudeboost -e "ANTHROPIC_API_KEY=$API_KEY" -- $PYTHON_PATH -m claudeboost_mcp

Display: ✅ MCP server registered. Restart Claude Code to activate. STOP here.

If `$ARGUMENTS` is `--login` or `login`: Run the login command using Bash:

claudeboost-mcp --login

This will prompt for email and password directly in the terminal. No browser needed. STOP here after the command completes. Do NOT call any MCP tools.

If `$ARGUMENTS` is `--logout` or `logout`: Delete the auth file using Bash:

rm -f ~/.claudeboost/auth.json

Then display: ✅ **Logged out of ClaudeBoost.** Run /boost --login to sign in again. STOP here.

If `$ARGUMENTS` is `--status` or `status`: Call the boost_status MCP tool. Display:

⚡ **ClaudeBoost Status**

| | |
|---|---|
| **Account** | {email or "Not signed in"} |
| **User ID** | {user_id or "—"} |
| **Boost Level** | {settings.boost_level} |
| **Auto-Boost** | {settings.auto_boost ? "ON" : "OFF"} |
| **Streak** | {streak.streak}-day · {streak.total_boosts} total |
| **Connected since** | {created_at or "—"} |

{if not authenticated: "Run `/boost --login` to sign in."}

STOP here.

If `$ARGUMENTS` is `--help` or `help`: Invoke the /boost-help skill instead. STOP here.

Otherwise (normal boost):

1. Call the boost_prompt MCP tool with the user's prompt: $ARGUMENTS

1. Check if the response has "skipped": true. If so, display:

✅ **Your prompt scores {original_score.total}/30 — already well-structured!**

No boost needed. Strong dimensions: {list strong_dimensions as comma-separated labels}.

Proceeding with your original prompt.

Then execute the original prompt directly. Call log_boost with {"original": original, "boosted": original, "domain": domain, "chosen": "skipped"}. STOP.

1. If NOT skipped, parse: domain, original, boosted, level, original_score, boosted_score, improvement, and improvements_added

1. Display the FULL comparison using this EXACT markdown format:

⚡ **CLAUDEBOOST** · `{domain}` · Level: `{level}`

📈 **Score: {original_score.total}/30 → {boosted_score.total}/30** (+{improvement})
🔧 **Boost added:** {join improvements_added with ", "}

---

> 📝 **Original:** {original}

---

### ✨ Boosted Prompt

{paste the COMPLETE boosted prompt here — every single line, no truncation}

---

📊 **Score Breakdown:**

| Dimension | Before | After |
|-----------|--------|-------|
| Specificity | {original_score.dimensions.specificity} | {boosted_score.dimensions.specificity} |
| Verification | {original_score.dimensions.verification} | {boosted_score.dimensions.verification} |
| Context | {original_score.dimensions.context} | {boosted_score.dimensions.context} |
| Constraints | {original_score.dimensions.constraints} | {boosted_score.dimensions.constraints} |
| Structure | {original_score.dimensions.structure} | {boosted_score.dimensions.structure} |
| Output | {original_score.dimensions.output_definition} | {boosted_score.dimensions.output_definition} |

---

{if streak.streak > 0: show "🔥 **{streak.streak}-day streak** · {streak.total_boosts} total boosts · {streak.today_boosts} today"}

1. AFTER displaying the full comparison above, present the choice using AskUserQuestion with these EXACT settings:

• question: "What would you like to do?"
• header: "ClaudeBoost"
• Option 1:
• label: "✨ Use boosted prompt (Recommended)"
• description: "Execute the enhanced version shown above"
• Option 2:
• label: "📝 Add notes & refine"
• description: "Give feedback to refine the boosted prompt"
• Option 3:
• label: "🚫 Keep original"
• description: "Ignore the boost and use your original prompt"
• Do NOT use previews on any option — the full prompts are already displayed above

1. IMPORTANT: Track the "current_boosted" variable. Initially it equals the boosted prompt from step 2. If the user refines it, update current_boosted to the refined version each time.

1. Based on the user's choice:

• If "Use boosted prompt":

1. Call log_boost MCP tool with: {"original": original, "boosted": current_boosted, "domain": domain, "chosen": "boosted"}

— current_boosted is the LATEST version (may have been refined multiple times) — Do NOT pass the first generated version if it was refined — Do NOT pass original_score/boosted_score — log_boost will re-score the final text

1. Then execute current_boosted as the user's new task. Do NOT mention ClaudeBoost again.

• If "Add notes & refine":

1. Output ONLY: 📝 **Type your notes below** (e.g. "use PyTorch instead of sklearn", "change 95% to 90%")
2. STOP and wait for user input.
3. When the user replies, apply their notes to current_boosted to create a new refined version.
4. Update current_boosted to this new refined version.
5. Display the refined version using the same markdown format (step 3) and present the choice modal again (step 4).
6. Repeat until user picks "Use boosted" or "Keep original".

• If "Keep original":

1. Call log_boost MCP tool with: {"original": original, "boosted": original, "domain": domain, "chosen": "original"}
2. Execute the original prompt as the user's task.

• If "Other":

1. Call log_boost MCP tool with: {"original": original, "boosted": user_typed_text, "domain": domain, "chosen": "refined"}
2. Execute whatever the user typed.

CRITICAL: Only call `log_boost` ONCE, after the user's FINAL choice. The `boosted` field in `log_boost` must be the FINAL text the user accepted — after ALL refinements, not the first generated version.

Important

• Do NOT skip the AskUserQuestion step. Always show the modal.
• Do NOT truncate or summarize the boosted prompt.
• Do NOT add extra commentary between the comparison and the modal.
• If the MCP tool is unavailable, tell the user: "ClaudeBoost MCP server is not connected. Run /mcp to check status."

Do NOT fall back to manual enhancement.

Authentication

If the MCP tool returns a JSON response with "error": "auth_required", run:

claudeboost-mcp --login

Then display: Run /boost again after signing in. Do NOT enhance manually. Do NOT retry. Wait for the user.