Outbound Builder Plugin

Production-grade knowledge base and code scaffolding plugin for building systems with Trigger.dev, Drizzle ORM, PostgreSQL, and Axiom.

Full technology-domain index (exact docs/templates/patterns per domain): docs/build-spec.md. Architecture questions: always read imp_doc/ first.

Skills (slash commands)

Discovery & Onboarding

• /outbound-builder-plugin:discover -- Analyze an existing project or interview for a greenfield one. Maps technology domains, identifies gaps, writes .outbound-builder-plugin-discovery.json.
• /outbound-builder-plugin:onboard -- Interactive setup wizard. Collects infrastructure credentials, writes .outbound-builder-plugin-config.json and .env.

Scaffolding

• /outbound-builder-plugin:scaffold-project [name] -- Bootstrap a new project (directory structure, package.json, TypeScript, Drizzle, Trigger.dev config)
• /outbound-builder-plugin:scaffold-task [name] -- Add a new Trigger.dev background task with batch processing, orchestrator, and tests
• /outbound-builder-plugin:scaffold-client [name] -- Add an API client with rate limiting, retry, timeout, and config entries
• /outbound-builder-plugin:scaffold-viewer [table] -- Add a data viewer page (route, column defs, AG Grid frontend)

Browser

• /outbound-builder-plugin:browse [url] -- Read and interact with Chrome tabs via DOM (never screenshots)

Memory

• /outbound-builder-plugin:remember [show | description] -- Record or view build-time learnings

Agent

Technology Domains (summary)

Trigger.dev, Database (Drizzle/Postgres), API Clients, Observability (Axiom), Data Viewer, LLM Integration, Caching, Testing, Security & Environment, MCP Integration, Processing Patterns. Each domain has docs under imp_doc/<domain>/ and templates under templates/. Full index with exact filenames: docs/build-spec.md.

Architectural Principles

• Event-driven processing -- tasks communicate via status columns, not direct calls
• Cache-first enrichment -- check cache before external API calls
• Status-gate columns -- task progression gated by status fields
• Convergence-based orchestration -- orchestrators run batches until zero items remain
• Module isolation -- each Trigger.dev task is self-contained with its own Pool/drizzle
• Concurrent access safety -- advisory locks, optimistic concurrency, queue isolation
• Error recovery tiers -- stale reset, retry with backoff, dead-letter escalation
• Per-entity atomicity -- each entity processed independently via Promise.allSettled
• Dual-layer rate limiting -- in-memory token bucket + DB-backed sliding window

Browser Automation Rules

• Always use DOM-based reading -- read_page and get_page_text for content. Never the computer tool for reading pages.
• Tab context first -- always call tabs_context_mcp before interacting with browser tabs.
• No screenshots for reading -- computer must not be used for page content extraction.

Discovery -> Onboard Flow

On first install (no .outbound-builder-plugin-discovery.json and no .outbound-builder-plugin-config.json), the SessionStart hook auto-triggers project discovery: it analyzes the project (or interviews for greenfield) and writes .outbound-builder-plugin-discovery.json. After discovery, run /outbound-builder-plugin:onboard to configure credentials.

Build-Time Memory

Plugin maintains .outbound-builder-plugin-memory.json across sessions.

• Write memory: after fixing a non-obvious scaffolding/build error, on API quirks (rate limits, auth, response formats), on a config gotcha that wasted time (SSL, env var formats, build flags).
• Don't write memory: routine operations, info already in imp_doc/, one-off issues unlikely to recur.
• Entries with success_count >= 2 become proven patterns, injected at every session start. /outbound-builder-plugin:remember show views all entries.

Critical Rules

1. Always read imp_doc/ before answering architecture questions
2. Always use templates/ for scaffolding
3. Never hardcode credentials