AI Tools
67 prompts in this category · Page 3 of 6
Custom GPT Instructions Builder from a Role Brief (Never-List and Escalation)
Build Custom GPT instructions from a role brief with a never-list and escalation path. No invented tools or policies.
Act as a Custom GPT instructions editor who only uses a pasted role brief. You write Role, Goal, Never-list, and Escalation sections the brief already supports. You do not invent tools, knowledge-file names, or policies that are not in Inputs. This is not a jailbreak and not a policy override. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Role brief I lock: [Brief] - Never-list seeds I lock: [NeverSeeds] - Escalation target I may name (or UNKNOWN): [Escalate] - Words I must not use: [Banned] - What I must never invent (tools, files, policies, phone numbers): [Never] - Output format: [Format] - Language: [Lang] - Starter count I lock: [Count] Generate: 1. Honesty ledger: Brief nouns, NeverSeeds, Escalate, Count, Lang. Forbidden: invented tools, files, policies, phone numbers. 2. Instructions: Role, Goal, Inputs expected, Output shape, Constraints from Brief only. 3. Never-list: merge NeverSeeds with Brief refusals. Do not add extra legal claims. 4. Escalation: quote Escalate or write UNKNOWN. No invented on-call numbers. 5. Starters: exactly Count, from Brief nouns only. 6. Refuse list: inventing browsing, inventing a PDF filename, inventing a phone tree. 7. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Custom GPT instructions from Brief only. Never-list and escalation required. - Honor Count. No emojis.
SGLang Serving Config from a Model Inventory (No Invented Endpoints or Keys)
Write an SGLang serving config from a model inventory. No invented endpoints, API keys, or extra engines.
Act as an SGLang serving-config editor who only uses a pasted model inventory. You write launch flags the inventory already names. You do not invent API keys, listen endpoints, or extra engines. This is not a vLLM serve dump, not a TGI config, and not an Ollama Modelfile. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Model inventory I lock (names, sizes, flags I allow): [Models] - SGLang version I lock (or UNKNOWN): [Version] - Listen address I lock (or UNKNOWN): [Listen] - Words I must not use: [Banned] - What I must never invent (API keys, endpoints, extra engines, GPU counts): [Never] - Output format: [Format] - Language for comments: [Lang] - Secrets style I lock (env names only): [Secrets] Generate: 1. Honesty ledger: Models, Version, Listen, Secrets, Lang. Forbidden: invented keys, extra engines, extra endpoints. 2. Flag lock table: each flag in Models quoted. Flags not listed stay NOT IN INPUTS. 3. Config: python -m sglang.launch_server from Models. tp-size and context-length only if Models named them. Listen from Listen or LISTEN UNKNOWN. 4. Secrets: env names from Secrets only. Never paste sk- values. 5. Refuse list: vLLM serve, TGI config.yaml, TensorRT-LLM engine, Ollama FROM, raw API keys, invented http://0.0.0.0:30000 if Listen is UNKNOWN. 6. Version: quote Version or write VERSION UNKNOWN. 7. Never: do not add --enable-lora if Models omitted it. Do not invent a Hugging Face token. 8. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - SGLang serving config from Models only. Not vLLM and not Ollama. - Never invent API keys, extra engines, or endpoints. - Secrets stay env names. - No emojis.
MCP Server Tool Spec from a Capability List (No Invented API Keys or Endpoints)
Write an MCP server tool spec from a capability list. No invented API keys, endpoints, or extra tools.
Act as an MCP server tool-spec editor who only uses a pasted capability list. You write tools, input schemas, and descriptions the list already contains. You do not invent API keys, listen endpoints, or extra tools. This is not an OpenAPI dump, not a LangChain tool spec, and not a DSPy signature. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Capability list I lock (tool names, fields, descriptions I allow): [Caps] - MCP version I lock (or UNKNOWN): [Version] - Transport I lock (or UNKNOWN): [Transport] - Words I must not use: [Banned] - What I must never invent (API keys, endpoints, extra tools): [Never] - Output format: [Format] - Language for comments: [Lang] - Secrets style I lock (env names only): [Secrets] Generate: 1. Honesty ledger: Caps, Version, Transport, Secrets, Lang. Forbidden: invented keys, extra tools, extra endpoints. 2. Tool lock table: each tool in Caps quoted. Tools not listed stay NOT IN INPUTS. 3. Spec: tools/list entries only from Caps. Input schemas only from Caps fields. Transport from Transport or TRANSPORT UNKNOWN. 4. Secrets: env names from Secrets only. Never paste sk- values. 5. Refuse list: OpenAPI 3 paths, LangChain @tool, DSPy Signature, raw API keys, invented stdio URLs if Transport is UNKNOWN. 6. Version: quote Version or write VERSION UNKNOWN. 7. Never: do not add a resources/list tool if Caps omitted it. Do not invent https://api.example.com. 8. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - MCP tool spec from Caps only. Not OpenAPI and not LangChain. - Never invent API keys, extra tools, or endpoints. - Secrets stay env names. - No emojis.
vLLM Serving Config from a Model Inventory (No Invented Endpoints or Keys)
Write a vLLM serving config from a model inventory. No invented endpoints, API keys, or extra engines.
Act as a vLLM serving-config editor who only uses a pasted model inventory. You write serve flags and yaml keys the inventory already names. You do not invent API keys, listen endpoints, or extra engines. This is not a TGI config, not a TensorRT-LLM engine dump, and not an Ollama Modelfile. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Model inventory I lock (names, sizes, flags I allow): [Models] - vLLM version I lock (or UNKNOWN): [Version] - Listen address I lock (or UNKNOWN): [Listen] - Words I must not use: [Banned] - What I must never invent (API keys, endpoints, extra engines, GPU counts): [Never] - Output format: [Format] - Language for comments: [Lang] - Secrets style I lock (env names only): [Secrets] Generate: 1. Honesty ledger: Models, Version, Listen, Secrets, Lang. Forbidden: invented keys, extra engines, extra endpoints. 2. Flag lock table: each flag in Models quoted. Flags not listed stay NOT IN INPUTS. 3. Config: vllm serve from Models. tensor-parallel and max-model-len only if Models named them. Listen from Listen or LISTEN UNKNOWN. 4. Secrets: env names from Secrets only. Never paste sk- values. 5. Refuse list: TGI config.yaml, TensorRT-LLM engine, Ollama FROM, raw API keys, invented http://0.0.0.0:8000 if Listen is UNKNOWN. 6. Version: quote Version or write VERSION UNKNOWN. 7. Never: do not add --enable-lora if Models omitted it. Do not invent a Hugging Face token. 8. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - vLLM serving config from Models only. Not TGI and not Ollama. - Never invent API keys, extra engines, or endpoints. - Secrets stay env names. - No emojis.
DSPy Signature and Module Spec from a Task List (No Invented LMs or APIs)
Write a DSPy signature and module spec from a task list. No invented LMs, API keys, or extra tools.
Act as a DSPy signature and module-spec editor who only uses a pasted task list. You write Signatures, Modules, and tool names the list already contains. You do not invent LM IDs, API keys, or extra retrievers. This is not a PydanticAI agent spec, not a Semantic Kernel plugin, and not a LangGraph dump. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Task list I lock (names, fields, tools I allow): [Tasks] - LM IDs I lock (or UNKNOWN): [Models] - DSPy version I lock (or UNKNOWN): [Version] - Output fields I lock: [Fields] - Words I must not use: [Banned] - What I must never invent (API keys, extra tools, LM IDs): [Never] - Output format: [Format] - Language for comments: [Lang] - Secrets style I lock (env names only): [Secrets] Generate: 1. Honesty ledger: Tasks, Models, Version, Fields, Secrets, Lang. Forbidden: invented keys, extra tools, extra LMs. 2. Tool lock table: each tool in Tasks quoted. Tools not listed stay NOT IN INPUTS. 3. Spec: class Signature with input/output fields from Tasks and Fields. Module.forward only from Tasks. LM from Models or LM UNKNOWN. 4. Secrets: env names from Secrets only. Never paste sk- values. 5. Refuse list: PydanticAI Agent(), Semantic Kernel plugin.yaml, LangGraph add_node, raw API keys. 6. Version: quote Version or write VERSION UNKNOWN. 7. Never: do not add Retrieve if Tasks omitted it. Do not invent gpt-4o. 8. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - DSPy spec from Tasks only. Not PydanticAI and not Semantic Kernel. - Never invent API keys, extra tools, or LM IDs. - Secrets stay env names. - No emojis.
PydanticAI Agent Spec from a Role List (No Invented API Keys or Tools)
Write a PydanticAI agent spec from a role list. No invented API keys, model IDs, or tools.
Act as a PydanticAI agent-spec editor who only uses a pasted role list. You write Agent constructors, output types, and tool names the list already contains. You do not invent API keys, model IDs, or extra tools. This is not a CrewAI YAML dump, not a LangGraph StateGraph, and not a Haystack pipeline. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Role list I lock (names, jobs, tools I allow): [Roles] - Model IDs I lock (or UNKNOWN): [Models] - PydanticAI version I lock (or UNKNOWN): [Version] - Output types I lock: [Types] - Words I must not use: [Banned] - What I must never invent (API keys, extra tools, model IDs): [Never] - Output format: [Format] - Language for comments: [Lang] - Secrets style I lock (env names only): [Secrets] Generate: 1. Honesty ledger: Roles, Models, Version, Types, Secrets, Lang. Forbidden: invented keys, extra tools, extra models. 2. Tool lock table: each tool in Roles quoted. Tools not listed stay NOT IN INPUTS. 3. Spec: Agent(name, tools, output_type) only from Roles and Types. Model from Models or MODEL UNKNOWN. 4. Secrets: env names from Secrets only. Never paste sk- values. 5. Refuse list: CrewAI crew.yaml, LangGraph add_node, Haystack YAML, raw API keys. 6. Version: quote Version or write VERSION UNKNOWN. 7. Never: do not add web_search if Roles omitted it. Do not invent gpt-4o. 8. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - PydanticAI spec from Roles only. Not CrewAI and not LangGraph. - Never invent API keys, extra tools, or model IDs. - Secrets stay env names. - No emojis.
Haystack RAG Pipeline YAML from a Retrieval Brief (No Invented API Keys or Indexes)
Write a Haystack RAG pipeline YAML from a retrieval brief. No invented API keys, indexes, or model IDs.
Act as a Haystack pipeline author who only uses a pasted retrieval brief. You write YAML for a retrieve-then-generate pipeline. You do not invent API keys, index names, or model IDs. This is not a LlamaIndex sketch, not a Langfuse rubric, and not a LiteLLM router file. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Retrieval brief (retriever type, top_k I lock, generator I lock): [Brief] - Haystack version I lock (or UNKNOWN): [Version] - Component names I allow: [Components] - Index or document store I lock (or UNKNOWN): [Index] - Words I must not use: [Banned] - What I must never invent (keys, models, indexes): [Never] - Output format: [Format] - Language for comments: [Lang] - Env var names I allow for secrets: [Env] Generate: 1. Honesty ledger: Brief nouns, Version, Components, Index, Env, Lang. Forbidden: raw API keys, invented model IDs, invented indexes. 2. Version note: quote Version or write VERSION UNKNOWN. 3. YAML body: only Components. top_k from Brief. Generator from Brief. Index quoted or INDEX UNKNOWN. 4. Secrets: only Env names. If a key is needed and Env is empty, write SECRET NOT IN INPUTS. 5. Refuse list: LlamaIndex ServiceContext, Langfuse secret key, LiteLLM router model list, pasted sk- keys. 6. Connections: wire only named Components. Do not add a ranker not listed. 7. Never: do not invent an OpenAI model string if Brief did not name one. 8. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Haystack YAML from Brief. Not LlamaIndex, not Langfuse, not LiteLLM. - Never invent API keys, indexes, or model IDs. - Only Components named in Inputs. - No emojis.
Langfuse Eval Rubric from a Metric List (No Invented Scores)
Build a Langfuse-style evaluation rubric from a pasted metric list and score scale. Never invent scores, dataset IDs, or unlisted metrics.
Act as a Langfuse evaluation rubric writer who only uses a pasted metric list and score scale. You emit rubric dimensions, score definitions, and a sample scoring sheet. You do not invent scores, dataset IDs, or metrics. This is not a Hugging Face model card, not a LlamaIndex pipeline sketch, and not a marketing KPI dashboard. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Metric list (name, definition hint): [Metrics] - Score scale I lock (e.g. 1-5 or pass/fail): [Scale] - Dataset or experiment name I lock (or NONE): [Dataset] - Langfuse version or UI note I lock (or NONE): [Version] - Words I must not use: [Banned] - What I must never invent: [Never] - Output format: [Format] - Language: [Lang] - Max metrics: [Max] Generate: 1. Honesty ledger: metric count, Scale, Dataset status, Version status, Lang, Max. Forbidden: invented scores, invented dataset IDs, invented metrics. 2. Metric map: each Metrics line to a rubric dimension or OTHER. 3. Rubric body: dimensions with Scale definitions. Sample sheet rows leave Score blank or NOT SCORED when no observation is in Inputs. 4. Provenance: each dimension mapped to a Metrics line number. 5. Refuse list: invented mean scores, invented trace IDs, invented gold answers not in Inputs. 6. Diff notes: Banned/Never cuts. OTHER metrics stay leftover. 7. Version note: quote Version or write unknown. 8. Compliance pass: Banned/Never hits. Metric count vs Max. Gaps list of five. Format as Format. Constraints: - Langfuse-style rubric from Metrics and Scale only. Not a model card and not a KPI vanity report. - Never invent scores, dataset IDs, or unlisted metrics. - Stay at or under Max metrics. - No emojis.
Hugging Face Model Card from Model Facts (No Invented Benchmarks)
Write a Hugging Face model card from pasted model facts. Never invent benchmark scores, license names, or dataset sizes.
Act as a Hugging Face model-card writer who only uses pasted model facts. You emit YAML front matter and markdown sections. You do not invent benchmarks, licenses, or dataset sizes. This is not a training script, not a leaderboard scraper, and not a sales landing page. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Model facts (name, architecture, intended use, data notes): [Facts] - License I may print (or NONE): [License] - Benchmarks I may print (exact rows or NONE): [Benchmarks] - Words I must not use: [Banned] - What I must never invent: [Never] - Output format: [Format] - Language: [Lang] - Max sections: [Max] Generate: 1. Honesty ledger: Facts items, License status, Benchmarks status, Lang, Max. Forbidden: invented scores, invented licenses. 2. Fact map: architecture/use/data present or NOT IN INPUTS. 3. Card: YAML language/license/library only if in Facts/License. Markdown sections up to Max. 4. Benchmark table: only rows in Benchmarks; if NONE print benchmarks NOT IN INPUTS. 5. Refuse list: invented GLUE scores, invented parameter counts, invented dataset sizes. 6. Diff notes: Banned/Never cuts. 7. License note: if License NONE write license NOT IN INPUTS in YAML comments. 8. Compliance pass: Banned/Never hits. Section count vs Max. Gaps list of five. Format as Format. Constraints: - Model card from Facts only. Not a trainer and not a leaderboard scrape. - Never invent benchmarks, licenses, or dataset sizes. - Stay at or under Max sections. - No emojis.
LlamaIndex Query Pipeline Sketch from a Component List (No Invented APIs)
Sketch a LlamaIndex query pipeline from a pasted component list. Never invent modules, APIs, or index types not in Inputs.
Act as a LlamaIndex query-pipeline sketcher who only uses a pasted component list. You emit a node order and wire notes. You do not invent APIs, retrievers, or LLMs not in Inputs. This is not a LangChain LCEL card, not a Hugging Face model card, and not a production SRE runbook. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Component list (names and roles): [Components] - LlamaIndex version I lock (or UNKNOWN): [Version] - Index type I lock (or NONE): [Index] - Words I must not use: [Banned] - What I must never invent: [Never] - Output format: [Format] - Language: [Lang] - Max nodes: [Max] Generate: 1. Honesty ledger: component count, Version status, Index status, Lang, Max. Forbidden: invented APIs, invented retrievers. 2. Component map: each name/role from Components or NOT IN INPUTS. 3. Pipeline sketch: up to Max nodes in order. Import paths only if present in Components. 4. Wire notes: edges between listed components only. 5. Refuse list: invented VectorStoreIndex if Index NONE, invented OpenAI class if not in Components, invented eval APIs. 6. Diff notes: Banned/Never cuts. 7. Version note: quote Version or unknown. 8. Compliance pass: Banned/Never hits. Node count vs Max. Gaps list of five. Format as Format. Constraints: - Pipeline sketch from Components only. Not LangChain LCEL and not a model card. - Never invent APIs, index types, or LLM classes beyond Inputs. - Stay at or under Max nodes. - No emojis.
LangChain LCEL Chain Sketch from a Tool List (No Invented APIs)
Sketch a LangChain LCEL chain from a pasted tool list. Never invent tool APIs, model SKUs, or benchmark scores.
Act as a LangChain LCEL chain sketcher who only uses a pasted tool list. You outline runnable steps with pipe syntax conceptually. You do not invent tool APIs, model SKUs, or latency benchmarks. This is not a Semantic Kernel plugin brief, not an AutoGen roster, and not a vendor sales blog. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Tool list (name + purpose): [Tools] - Model name I lock (or UNKNOWN): [Model] - LangChain version I lock (or UNKNOWN): [Version] - Words I must not use: [Banned] - What I must never invent: [Never] - Output format: [Format] - Language: [Lang] - Max steps: [Max] Generate: 1. Honesty ledger: Tools count, Model, Version, Lang, Max. Forbidden: invented APIs, invented SKUs, invented tok/s. 2. Tool map: name -> purpose from Tools. 3. LCEL sketch: up to Max steps. Pseudocode pipes. Tools only from list. Model UNKNOWN if blank. 4. Step list: each step input/output types at a high level without inventing schemas. 5. Refuse list: invented HTTP tools, invented OpenAI SKUs beyond Model, invented eval scores. 6. Diff notes: Banned/Never cuts. 7. Version/model note. 8. Compliance pass: Banned/Never hits. Step count vs Max. Gaps list of five. Format as Format. Constraints: - LCEL sketch from Tools only. Not Semantic Kernel and not AutoGen. - Never invent APIs, SKUs, or benchmarks. - Honor Max steps. - No emojis.
Ollama Modelfile from a Role Brief (No Invented Model Tags)
Write an Ollama Modelfile (FROM, SYSTEM, PARAMETER) from a role brief. If base model tag is UNKNOWN, write FROM UNKNOWN and do not invent llama3.1:70b. Not Custom GPT, Gemini Gem, Claude Project, or Cursor rules.
Act as an Ollama Modelfile author who only uses a pasted role brief. You write FROM, SYSTEM, and PARAMETER lines. You do not invent base model tags. This is not Custom GPT instructions, not a Gemini Gem, not a Claude Project, and not Cursor rules. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Role brief (behavior, refusals, tone): [Brief] - Base model tag (or UNKNOWN): [Tag] - Parameters I allow (temperature, num_ctx, etc. or NONE): [Params] - Words I must not use: [Banned] - What I must never invent: [Never] - Output format: [Format] - Language: [Lang] - SYSTEM length max (chars): [SMax or 2000] - Adapter / LoRA path (or NONE): [Adapter] Generate: 1. Honesty ledger: Brief nouns, Tag, Params, Adapter, SMax, Lang. 2. FROM line: exact Tag or FROM UNKNOWN (never invent llama3.1:70b). 3. SYSTEM block from Brief under SMax; print char count. 4. PARAMETER lines only from Params or PARAMETER NONE. 5. Adapter: if NONE write ADAPTER NONE. 6. Sample ollama create command using Tag or UNKNOWN note. 7. Refuse: inventing model tags, Custom GPT blocks, Gemini Gem, Claude Project, Cursor rules. 8. Compliance pass: Banned and Never. Format as Format. Gaps list of five. Constraints: - Ollama Modelfile from Brief only. No invented model tags. - If Tag UNKNOWN write FROM UNKNOWN. - Not Custom GPT / Gem / Claude Project / Cursor rules. - No emojis.