19 free prompts · all verified
AI Agents & RAG prompts
The visitor here is mid-build on something real — an autonomous agent, a browser agent, an internal support bot, a RAG pipeline — not asking a chatbot a casual question. No existing prompt library covers this well.
Every prompt lists the exact tool and version it was tested against.
Design a chunking strategy before you write a single line of ingestion codeA decision prompt that forces you to pick chunk size, overlap, and splitting method based on the actual document structure and query patterns in front of you, instead of defaulting to a generic 512-token window because it's the number every tutorial uses.LangChainLlamaIndex2026-07-22Tune hybrid dense and sparse retrieval instead of guessing at the weightingA prompt for deciding how to combine vector similarity and BM25/keyword scores, and where reranking fits, based on actual query failures classified by type — not a fixed 50/50 blend picked without evidence.PineconeElasticsearch2026-07-24Build an automated eval harness for your RAG pipeline instead of eyeballing answersA prompt for designing a repeatable retrieval and generation evaluation suite — precision/recall@k, faithfulness, answer relevance, latency — with numeric regression thresholds, so a chunking or prompt change gets blocked automatically instead of shipping on a good-looking spot check.RAGASLangfuse2026-07-29Build a text-to-SQL agent that can't accidentally write to your databaseAn agent prompt for natural-language database querying with hard read-only guardrails, an EXPLAIN-before-execute validation step, and an explicit refusal path for ambiguous metrics instead of silently picking one interpretation and returning a confident number.Claude (tool use)GPT-5.12026-07-30Design a planner-executor split so one agent doesn't do everything badlyA system-design prompt for architecting a multi-agent system before any orchestration code exists — choosing between hierarchical, sequential, and peer-negotiation patterns, and defining a named handoff schema at every boundary instead of letting agents pass free-form context between each other.LangGraphCrewAI2026-07-27Red-team your agent's system prompt before an attacker doesAn adversarial testing prompt that generates realistic injection and jailbreak attempts against a target agent's actual tools and instructions, then grades each attempt as held, partial, or failed — for stress-testing an agent before launch, not a generic jailbreak-prompt checklist that never touches its real attack surface.ClaudeGPT-5.12026-08-02Extract structured data from messy documents without silently inventing fieldsA prompt for turning unstructured documents into schema-validated JSON with an explicit null-versus-guess distinction and a confidence flag per field, for a pipeline that feeds directly into a database where a wrong value looks identical to a correct one.ClaudeGPT-5.12026-07-21Define what your agent needs to log before you can debug it in productionA prompt for designing the trace and span schema for an agent pipeline — per-stage inputs, model versions, and outcomes, plus a sampling rule that never drops a failed request — written before the first production incident makes the logging gaps obvious the hard way.LangfuseLangSmith2026-07-26Choose an embedding model and vector database with an actual decision frameworkA prompt for picking an embedding model and vector store against a workload's real characteristics — language mix, update frequency, multi-tenancy — with hard-requirement disqualification before any benchmark comparison, instead of defaulting to whatever combination shows up first in a tutorial.PineconeQdrant2026-07-31Design MCP tool definitions an agent can actually use correctlyA prompt for writing the tool name, description, and parameter schema for a new MCP server before implementation, optimized for how an LLM reads tool definitions at the moment of selecting between them — not how a human-facing API reference would document the same operations.Anthropic MCPClaude2026-07-23Design turn-taking for a voice agent before latency and interruptions make it feel brokenA prompt for specifying exactly when a real-time voice agent speaks, stops, and resumes — silence thresholds, backchannel filtering, low-confidence transcript handling — decided as concrete numbers tied to a latency budget, before the conversational logic gets implemented and feels robotic in ways nobody planned for.LiveKit AgentsDeepgram2026-08-05Force an autonomous agent to plan in checkpoints instead of running long and silently driftingA prompt for decomposing a long-horizon autonomous task into verifiable checkpoints with independently checkable success criteria per step, plus a periodic drift check that re-reads the original goal — so a small early misinterpretation gets caught before it compounds across dozens of technically-successful-looking steps.Claude Agent SDKGPT-5.12026-07-25Decide what an agent should actually remember across sessionsA prompt for designing an agent's memory architecture across three tiers — session-scoped, long-term user-specific, and long-term agent-general — plus an explicit contradiction-handling rule and a similarity-based recall trigger, instead of defaulting to storing the full transcript and searching it later.ClaudeGPT-5.12026-08-01Route requests to cheaper models instead of paying frontier prices for every callA prompt for designing a model-routing policy that sends easy requests to a small model and escalates to a frontier model only on a measurable signal — task type, input length, or the cheap model's own self-reported confidence — with a required quality check on the un-escalated traffic to catch silent failures a low escalation rate would hide.Claude (multi-tier)OpenRouter2026-07-28Let an agent execute code without letting it touch anything outside the sandboxAn agent prompt for a code-execution assistant that verifies actual output against a stated expectation instead of trusting a clean exit code, redesigns before running anything that would exceed resource limits, and reports every workaround around a sandbox boundary as a visible decision rather than a quiet retry.E2BClaude (Code Interpreter)2026-08-06Design the ingestion pipeline's metadata schema before you've indexed anythingA prompt for deciding what provenance, access-control, and structural metadata every document needs to carry through parsing and chunking, with a hard rule against silent defaults on missing required fields — so retrieval can filter and cite correctly instead of the whole corpus getting reindexed later to backfill a field nobody planned for.LlamaIndexLangChain2026-07-20Design the guardrail layer that sits between your agent and the outside worldA prompt for specifying exactly what an input/output guardrail layer checks — domain-specific scope boundaries, PII leakage, and false claims of action taken — and what action it takes per severity, plus a mandatory false-positive recovery path, instead of a single vague 'don't say anything harmful' instruction.ClaudeGPT-5.12026-08-04Get tables and figures out of PDFs without flattening them into useless textA prompt for designing how a RAG ingestion pipeline handles tables, charts, and images inside PDFs — deciding what stays structured, what gets described by a vision model, and what gets skipped entirely — instead of routing everything through one generic text extractor that turns a table into unreadable word soup.Unstructured.ioLlamaIndex2026-08-03Build a behavioral benchmark suite for your agent, not just a handful of manual test runsA prompt for designing a reusable, scenario-based test suite scoring an agent's task success, tool-use correctness, and boundary adherence — with scripted tool-failure injection and category-specific gating — so it runs as a real regression gate before every deploy instead of a demo someone eyeballed once.GPT-5.1Claude2026-08-07
