Search-Gateway

Implement and maintain CRMind user API key issuance, verification, revocation, and frontend management flows. Use when working on /auth/api-keys endpoints, middleware auth fallback, key rotation UX, or API key documentation.

Implement the ARQ background worker process for crawl queue processing, Server-Sent Events for streaming agent progress, multi-stage Docker builds, and the worker Docker Compose service. Use when the crawl queue needs a consumer, when adding real-time agent progress to the frontend, or when optimising the Docker image size. Keywords: ARQ, worker, background job, crawl queue, SSE, Server-Sent Events, agent streaming, Docker multi-stage, worker process, EventSource.

Format LLM synthesis output into structured CRMind responses with traceable source citations. Use this skill when attaching source links to facts, building the canonical JSON response shape, formatting citation cards, or ensuring every claim in the response is backed by a source URL with metadata. Keywords: citation, source link, structured response, fact attribution, evidence, response formatting, citation card, source traceability, audit trail.

Manage database schema migrations with Alembic and handle embedding model versioning safely. Use when adding columns or tables, running ALTER statements, upgrading the embedding model, or configuring Supabase connection pooling. Keywords: Alembic, migration, ALTER TABLE, embed_model_id, embedding versioning, re-embed, rembed_chunks, Supabase, PgBouncer, schema change.

Configure and manage the free-tier deployment pipeline for CRMind using Vercel, Supabase, Render/Cloudflare, and GitHub Actions. Use this skill when setting up CI/CD, batch ingestion cron jobs, environment configuration, Supabase migrations, Vercel deployments, caching strategies, or cost optimisation for low-resource hosting. Keywords: deployment, Vercel, Supabase, Render, GitHub Actions, CRON, batch ingestion, free tier, CI/CD, environment variables, migrations.

Resolve a company name, person name, email domain, or account ID to a canonical entity record in the CRMind database. Use this skill when implementing fuzzy matching, deduplication, canonical ID assignment, alias handling, or any code that maps a raw input string to a structured entity. Keywords: entity resolution, canonical ID, company lookup, person lookup, fuzzy match, dedup, alias, domain match.

Implement retry logic, circuit breakers, graceful degradation, fallback responses, input sanitization, and exception hierarchies for CRMind. Use when writing any code that calls an LLM, embedding model, external HTTP, or database — and when building error handlers in FastAPI. Keywords: retry, circuit breaker, tenacity, graceful degradation, fallback, sanitize, prompt injection, exception hierarchy, LLMUnavailableError, degraded response.

Define and implement the FastAPI backend structure for CRMind. Use this skill when creating API routes, request/response models, middleware, database connections, dependency injection, error handling, or background task setup. Keywords: FastAPI, API routes, Pydantic, endpoint, async, database, dependency injection, middleware, background tasks, CORS, authentication, rate limiting.

Build complete production-ready frontends — website, web app, SaaS, landing page, dashboard, React app. Generates all pages (Homepage, About, Services, Contact, Login, Signup) plus Navbar and Footer with a full design system, theme, and typography.

Implement LangGraph-based tool-calling agent workflows for CRMind. Use this skill when building or modifying any agent: Lead Finder, Account Brief, CRM Enrichment, Research, or Ops/Debug. Covers StateGraph setup, node functions, conditional edges, tool binding, and structured output. Keywords: LangGraph, agent, StateGraph, tool calling, agent workflow, node, edge, conditional routing, agent state, lead finder, account brief, enrichment agent.

Add trace IDs, structured logging, request timing, circuit breaker state, and the /explain endpoint to CRMind. Use when adding any logging, tracing, metrics, health checks, or the debug panel API. Keywords: trace ID, structured logging, loguru, request timing, observability, X-Trace-ID, middleware, health endpoint, explain endpoint, log correlation.

Add pagination to all list endpoints, implement the PaginatedResponse generic, ETag caching, idempotency keys for batch enrich, OpenAPI tags, rate limiting with slowapi, and the required fields checklist for every API response. Use when creating or modifying any endpoint that returns a list, when adding rate limiting, or when validating response shapes. Keywords: pagination, PaginatedResponse, limit offset, total, has_more, ETag, rate limit, slowapi, idempotency, OpenAPI, response contract.

Implement the retrieval pipeline for CRMind: vector search, keyword search, metadata filtering, freshness ranking, trust scoring, and result merging. Use this skill when writing retrieval logic, chunk ranking, pgvector queries, similarity search, or any code that fetches context to ground LLM responses. Keywords: RAG, retrieval, vector search, pgvector, ranking, freshness, trust score, cosine similarity, hybrid search, chunk retrieval.

Describe what this skill does and when to use it. Include keywords that help agents identify relevant tasks.

Implement trust scoring, freshness scoring, and source authority ranking for CRMind. Use this skill when computing scores for retrieved chunks, source documents, entities, or signals. Covers exponential decay freshness, source type authority weights, cross-reference boosting, and score persistence. Keywords: trust score, freshness score, source authority, scoring, ranking, exponential decay, reliability, evidence quality, signal confidence.

Implement the async web crawling layer for CRMind using Playwright. Use this skill when writing page fetching, dynamic rendering, HTML extraction, robots.txt checking, crawl queue management, or any ingestion code that retrieves raw web content. Keywords: crawler, Playwright, web scraping, fetch page, extract text, crawl queue, robots.txt, async crawl, HTML to text, ingestion.