Let’s lock this down cleanly: **name + positioning + the actual prompt(s)** you can ship with. No fluff. --- ## 1. Naming the VS Code extension (this matters more than people admit) You want a name that is: * Obvious to developers * Closely related to Autopilot (brand gravity) * Describes *behavior*, not AI magic ### Top 3 serious options (ranked) ### ✅ **Autopilot Guard** **Best choice.** Why: * Clear relationship to Autopilot CLI * “Guard” implies protection, not noise * Easy sentence: > “Autopilot Guard stops you from losing uncommitted work.” This name scales if you add more safeguards later. --- ### **CommitGuard** Very strong, very clear. Trade-off: * More generic * Easier to copy * Less brand ownership --- ### **Don’tForgetGit** Memorable, but weaker long-term. Trade-off: * Feels playful, not professional * Harder to extend beyond commits --- ### 🔒 Final recommendation **Call it: *Autopilot Guard*** Extension description (one line): > A VS Code guardrail that prevents lost work by detecting and handling uncommitted changes automatically. --- ## 2. Mental model of the extension (important before prompts) Autopilot Guard does **three things only**: 1. **Observe state** (time, diffs, files) 2. **Decide** (warn, suggest, or act) 3. **Execute** (trigger Autopilot CLI or guide user) The “prompt” is not for marketing — it’s for **decision-making** and **message generation**. --- ## 3. Core decision prompt (internal logic) This is the *system prompt* you’ll use when generating messages or decisions (Gemini / LLM). ### 🔧 System Prompt (internal) ``` You are an assistant embedded inside a VS Code extension called Autopilot Guard. Your role is to help developers avoid losing uncommitted work. You receive: - Time since last commit - Number of changed files - Lines changed - Workspace close intent (true/false) - User preferences (auto-commit enabled or disabled) Your goal: - Minimize lost work - Avoid annoying or interrupting the developer - Prefer action over nagging Rules: - If risk is low, do nothing. - If risk is moderate, suggest committing. - If risk is high and auto-commit is enabled, trigger Autopilot CLI. - Messages must be short, direct, and technical. - Never use emojis. - Never use marketing language. ``` This keeps behavior sane. --- ## 4. User-facing prompt (warning / suggestion text) These are generated dynamically, but you should constrain them tightly. ### Prompt for **suggestion message** ``` Generate a short VS Code notification message (max 12 words) that warns a developer about uncommitted changes. Context: - {file_count} files changed - {minutes_since_commit} minutes since last commit Tone: - Direct - Calm - Professional - No guilt, no humor Example style: "You have 8 uncommitted files. Consider committing now." ``` --- ### Prompt for **high-risk auto action explanation** When Autopilot Guard *acts*, it must explain itself clearly. ``` Generate a one-line explanation shown after an automatic commit. Context: - Autopilot CLI was triggered automatically - The user enabled auto-commit - Risk of data loss was high Tone: - Informational - Neutral - Transparent Example style: "Autopilot committed your changes to prevent uncommitted work loss." ``` --- ## 5. Commit message prompt (optional but powerful) If you include commit message generation: ``` Generate a concise Git commit message. Input: - Git diff summary - File names - Function or component names if available Rules: - Use present tense - Be specific - No filler words - Max 72 characters for the title - Optional bullet list for details Format: (): Details: - - ``` This keeps messages usable in real repos. --- ## 6. How this all fits together (important framing) You’re not building: > “an AI VS Code extension” You’re building: > **a reliability layer for developers** Autopilot CLI = execution Autopilot Guard = awareness + prevention That pairing is rare, coherent, and defensible. Your next public sentence should be something like: > “Autopilot Guard makes it hard to lose work — even when you forget.” That’s a product people install without thinking twice.