App Blink exposes a suite of Language Model Tools directly to the editor's language model tool-calling runtime (vscode.lm.tools). Autonomous AI agents (such as GitHub Copilot Agent Mode, Claude Code, Cursor, and Google Antigravity) call these tools to respect workspace guardrails, enforce DRY policies, reuse existing monorepo components, and advance task queues.
🛠️ Tool Registry & Schemas
1. Guardrail Fetcher (`#appblink_vibecheck_guardrail`)
* **When to call**: At the **START** of any code-generation or code-modification task. * **What it does**: Returns the workspace DRY policy, active framework conventions, and configured VibeCheck rules filtered to the target file paths. * **Input Schema**: ```json { "filePaths": ["string[] (optional relative paths of files being modified)"], "taskDescription": "string (optional description of the task)" } ``` * **Output**: Formatted plain-text guardrail block that the agent must satisfy.2. Component Reuse Check (`#appblink_component_reuse_check`)
* **When to call**: **BEFORE** writing any new widget, card, button, dialog, snackbar, or UI element. * **What it does**: Searches `.appblink/codeindex.md` across shared packages (`flutter/packages/workern_widgets`, `angular/packages/components`, `spartan`) to locate existing components matching the description. * **Behavior**: * If a match is found, the agent **must use it** instead of writing redundant code. * If an app-level component exists, the tool suggests promoting it to a shared package. * **Input Schema**: ```json { "componentDescription": "string (e.g. 'rating stars widget')", "framework": "flutter | angular | both", "filePaths": ["string[] (optional)"] } ```3. Design Skill Router (`#appblink_design_skill_router`)
* **When to call**: At the start of visual design, UI styling, or UX tasks. * **What it does**: Analyzes the task description and returns the exact path to the appropriate design `SKILL.md` file: * `banner-design`: Ad banners, social covers, OG images * `brand`: Brand guidelines, typography, logo rules * `design-system`: Design tokens, CSS variables, spacing scales * `ui-styling`: Tailwind, Spartan UI, Shadcn/Radix components * `ui-ux-pro-max`: Full product UI/UX, accessibility, layout * **Input Schema**: ```json { "taskDescription": "string (required description of the UI task)", "filePaths": ["string[] (optional)"] } ```4. Vibe Check Signal (`#appblink_vibecheck_complete`)
* **When to call**: After the agent has run all required vibe checks for a task. * **What it does**: Signals the pass/fail outcome to the extension and records check summaries. * **Input Schema**: ```json { "taskId": "string (exact task ID)", "status": "passed | failed", "summary": "string (optional summary of checks and fixes)" } ```5. Task Completion Signal (`#appblink_task_complete`)
* **When to call**: When **ALL** implementation steps, tests, and vibe checks are finished. * **What it does**: Fires a completion signal that triggers App Blink to: 1. Stage working tree changes 2. Create a standardized Git commit 3. Mark the task complete in Firestore 4. Dispatch the next queued task * **Input Schema**: ```json { "taskId": "string (exact task ID)", "summary": "string (summary of changes made)" } ```🤖 Standard Agent Execution Loop
Autonomous agents in an App Blink workspace should follow this deterministic lifecycle:
graph TD
A[Start Task] --> B[Call #appblink_vibecheck_guardrail]
B --> C{Is Task UI/Design?}
C -->|Yes| D[Call #appblink_component_reuse_check]
D --> E[Call #appblink_design_skill_router]
C -->|No| F[Implement Logic]
E --> F
F --> G[Run Vibe Checks]
G --> H[Call #appblink_vibecheck_complete]
H --> I[Call #appblink_task_complete]
I --> J[Done & Advanced]