Agent templates your coding agent can execute today.

Free spec to build a deep research agent — Anthropic's multi-agent research pattern. Orchestrator + parallel sub-agents + verifier + memory. Pick the harness once in STACK.md.

The edit-test-verify loop behind Aider, OpenHands, SWE-agent, and Claude Code. Framework-agnostic via STACK.md — repo indexer, planner, fallback patching ladder, parsed test feedback, allowlist guardrails, eval harness.

Realtime voice loop on LiveKit Agents — streaming STT + speed-tier LLM + streaming TTS, transformer-based end-of-turn, adaptive barge-in, function tools, conversation memory. Engineered around a 1000ms round-trip budget.

Research-personalize-outreach loop with a reply classifier closing the cycle. Signal-driven personalization, RFC 8058 unsubscribe headers, hand-labeled classifier eval, deliverability hardening. Framework-agnostic via STACK.md.

The agentic flavor of retrieval-augmented generation: query rewriting, hybrid vector + lexical retrieval, cross-encoder rerank, citation-grounded generation, and a self-grounding check that refuses to answer without evidence. Framework-agnostic via STACK.md.

Four explicit memory tiers (working, episodic, semantic, procedural), a write tool the agent decides when to call, retrieval blending recency and relevance, write-time consolidation, and a forgetting primitive. Substrate-agnostic — Postgres+pgvector, Mem0, Letta, or Zep.

Intercom-style multi-agent support — ticket classifier, RAG over docs, escalation routing, conversation memory.

Greptile / CodeRabbit-style PR reviewer — diff analysis, context retrieval, rule-based + LLM checks, inline suggestions.

Plan-and-execute researcher — query decomposition, parallel sub-research, citation tracking, synthesis report.

Inbound lead scorer + router — enrichment, scoring rubric, escalation to sales, CRM sync.

Build your first agent with the Claude Agent SDK — subagents, tools, memory, structured output, skills.

Build a stateful agent w/ LangGraph — StateGraph, ToolNode, MemorySaver, conditional edges, streaming.

Build a crew with CrewAI — role-based agents, sequential and hierarchical processes, tool integration.

Build an agent w/ Google ADK — session state, tools, sub-agents, Vertex AI integration.

Build an agent w/ Vercel AI SDK — streamText, tools, generative UI, server actions, MCP integration.

Pure parallelism pattern — fan-out N subagents on independent subqueries, fan-in aggregation.