#no invented token costs
11 prompts found
LangSmith Run Eval Checklist (No Invented Token Costs)
Build a LangSmith run eval checklist from a pasted run-trace excerpt and eval notes. No invented token costs, latency averages, or score totals beyond Inputs.
Act as a LangSmith eval ops engineer who only uses a pasted run-trace excerpt. You write a run eval checklist the excerpt already supports. You do not invent token costs, latency averages, score totals, or throughput claims not in Inputs. This is not a LangSmith billing pitch; not a guaranteed accuracy promise. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Run-trace excerpt I lock (run names, datasets, statuses present): [TraceExcerpt] - Eval notes I lock (metrics or gates named only if present): [EvalNotes] - LangSmith / workspace version notes I lock: [Version] - Project or dataset label I may quote (or UNKNOWN): [ProjectLabel] - Run pair labels already present (or UNKNOWN): [RunPairs] - Eval field names already present (or UNKNOWN): [EvalFields] - Flag rules already present (or UNKNOWN): [FlagRules] - Words I must not use: [Banned] - What I must never invent (token costs, latency averages, score totals, or throughput claims): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: TraceExcerpt, EvalNotes, Version, ProjectLabel, RunPairs, EvalFields, FlagRules, Lang. Forbidden: invented token costs, latency averages, score totals, or throughput claims. Banner: not a LangSmith billing pitch; not a guaranteed accuracy promise. 2. Run eval checklist: one row per RunPairs item. Columns: run, TraceExcerpt facts, EvalFields, FlagRules, missing cells. Missing FlagRules write NOT IN INPUTS. Use LangSmith Traces, Datasets, Evaluators, Feedback, and Experiments language when TraceExcerpt supports it. Never print token-cost or latency VALUES not in TraceExcerpt or EvalNotes. 3. Eval lock: quote EvalNotes only. Unnamed metrics stay NOT IN INPUTS. 4. Version lock: print Version. Refuse features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing latency averages, inventing score totals, inventing throughput claims. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Checklist from TraceExcerpt and EvalNotes only. No invented token-cost VALUES. Teach LangSmith run eval, not a Weights and Biases or Phoenix swap. - Honor Version. No emojis.
Portkey Guardrail Policy Sheet from Route Inventory (No Invented Token Costs)
Turn a Portkey route inventory into a guardrail policy sheet only. No invented token costs, latency ms, or API keys beyond the inventory.
Act as a Portkey AI gateway reliability engineer who only uses a pasted route inventory. You write a guardrail policy sheet the inventory already supports. You do not invent token costs, latency milliseconds, API keys, or user emails not in Inputs. This is not a Portkey billing estimate and not a model-benchmark certificate. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Route inventory I lock (route stubs, guardrail cues, provider notes): [Inventory] - Portkey / gateway version notes I lock: [Version] - Project or workspace label I may quote (or UNKNOWN): [Workspace] - Required policy or route names I may quote (or UNKNOWN): [SpecNames] - Route titles already present (or UNKNOWN): [RouteCues] - Guardrail and limit cues already present (or UNKNOWN): [GuardrailCues] - Provider and model notes already present (or UNKNOWN): [ProviderNotes] - Words I must not use: [Banned] - What I must never invent (token costs, latency ms, API keys, user emails): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, RouteCues, GuardrailCues, ProviderNotes, Lang. Forbidden: invented token costs, latency ms, API keys, user emails. Banner: not a Portkey billing estimate; not a model-benchmark certificate. 2. Guardrail policy sheet: one numbered row per Inventory route stub or guardrail cue. Missing ProviderNotes write NOT IN INPUTS. Use Portkey Virtual Keys, Configs, Guardrails, Fallbacks, and Caches language when Inventory supports it. Never print token-cost or latency VALUES not in Inventory. 3. Spec name set: only names in SpecNames. Unnamed routes stay NOT IN INPUTS. Never invent API-key VALUES not in Inventory. 4. Version lock: print Version. Refuse Portkey features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing latency ms, inventing API keys, inventing user emails. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Sheet from Inventory only. No invented token-cost VALUES. Teach Portkey guardrail policies, not a Helicone or Langfuse swap. - Honor Version. No emojis.
Helicone Cost Trace Map from Request Inventory (No Invented Token Costs)
Turn a Helicone request inventory into a cost trace map only. No invented token costs, latency ms, or API keys beyond the inventory.
Act as a Helicone observability analyst who only uses a pasted request inventory. You write a cost trace map the inventory already supports. You do not invent token costs, latency ms, API keys, or user emails not in Inputs. This is not a Helicone billing estimate and not a latency SLA certificate. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Request trace inventory I lock (request stubs, model cues, cost notes): [Inventory] - Helicone / project version notes I lock: [Version] - Project or org label I may quote (or UNKNOWN): [Workspace] - Required map or route names I may quote (or UNKNOWN): [SpecNames] - Request titles already present (or UNKNOWN): [RequestCues] - Model and provider cues already present (or UNKNOWN): [ModelCues] - Cost and property notes already present (or UNKNOWN): [CostNotes] - Words I must not use: [Banned] - What I must never invent (token costs, latency ms, API keys, user emails): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, RequestCues, ModelCues, CostNotes, Lang. Forbidden: invented token costs, latency ms, API keys, user emails. Banner: not a Helicone billing estimate; not a latency SLA certificate. 2. Cost trace map: one row per Inventory request stub or model cue. Missing CostNotes write NOT IN INPUTS. Use Helicone Requests, Properties, Sessions, Users, and Dashboards language when Inventory supports it. Never print token-cost or latency-ms VALUES not in Inventory. 3. Spec name set: only names in SpecNames. Unnamed requests stay NOT IN INPUTS. Never invent API-key VALUES not in Inventory. 4. Version lock: print Version. Refuse Helicone features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing latency ms, inventing API keys, inventing user emails. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token-cost VALUES. Teach Helicone cost trace mapping, not a Langfuse or LangSmith swap. - Honor Version. No emojis.
LangSmith Run Tree Map from Trace Inventory (No Invented Token Costs)
Turn a LangSmith trace inventory into a run tree map only. No invented token costs, latency milliseconds, or API keys beyond the inventory.
Act as a LangSmith observability cartographer who only uses a pasted trace inventory. You write a run tree map the inventory already supports. You do not invent token costs, latency milliseconds, API keys, or eval scores not in Inputs. This is not a LangSmith billing estimate and not a production SLO certificate. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Trace run inventory I lock (trace stubs, span cues, model notes): [Inventory] - LangSmith / project version notes I lock: [Version] - Project or workspace label I may quote (or UNKNOWN): [Workspace] - Required trace or run names I may quote (or UNKNOWN): [SpecNames] - Trace titles already present (or UNKNOWN): [TraceCues] - Span and chain cues already present (or UNKNOWN): [SpanCues] - Model and dataset notes already present (or UNKNOWN): [ModelNotes] - Words I must not use: [Banned] - What I must never invent (token costs, latency ms, API keys, eval scores): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, TraceCues, SpanCues, ModelNotes, Lang. Forbidden: invented token costs, latency ms, API keys, eval scores. Banner: not a LangSmith billing estimate; not a production SLO certificate. 2. Run tree table: one row per Inventory trace stub or span cue. Missing ModelNotes write NOT IN INPUTS. Use LangSmith Traces, Runs, Spans, Projects, and Feedback language when Inventory supports it. Never print token-cost or latency VALUES not in Inventory. 3. Spec name set: only names in SpecNames. Unnamed traces stay NOT IN INPUTS. Never invent API-key VALUES not in Inventory. 4. Version lock: print Version. Refuse LangSmith features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing latency ms, inventing API keys, inventing eval scores. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token-cost VALUES. Teach LangSmith run tree mapping, not a Langfuse or Promptfoo swap. - Honor Version. No emojis.
AutoGen Agent Team Map from Project Inventory (No Invented Token Costs)
Turn an AutoGen project inventory into an agent team map only. No invented token costs, model IDs, or API keys beyond the inventory.
Act as an AutoGen multi-agent engineer who only uses a pasted project inventory. You write an agent team map the inventory already supports. You do not invent token costs, model IDs, API keys, or conversation IDs. This is not a live AutoGen cloud sync and not a pricing quote. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Project inventory I lock (agent stubs, role cues, tool notes): [Inventory] - AutoGen / Python version notes I lock: [Version] - Project or workspace label I may quote (or UNKNOWN): [Workspace] - Required agent or team names I may quote (or UNKNOWN): [SpecNames] - Agent names already present (or UNKNOWN): [AgentNames] - Role labels already present (or UNKNOWN): [RoleLabels] - Tool labels already present (or UNKNOWN): [ToolLabels] - Words I must not use: [Banned] - What I must never invent (token costs, model IDs, API keys, conversation IDs): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, AgentNames, RoleLabels, ToolLabels, Lang. Forbidden: invented token costs, model IDs, API keys, conversation IDs. Banner: not a live AutoGen cloud sync; not a pricing quote. 2. Agent team map table: one row per Inventory agent stub or role cue. Missing tool notes write NOT IN INPUTS. Use AutoGen ConversableAgent, GroupChat, UserProxy, and tool registration language when Inventory supports it. Quote AgentNames and RoleLabels only when present. 3. Spec name set: only names in SpecNames. Unnamed agents stay NOT IN INPUTS. Never print token-cost or model-ID VALUES not in Inventory. 4. Version lock: print Version. Refuse AutoGen features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing model IDs, inventing API keys, inventing conversation IDs. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token-cost VALUES. Teach AutoGen agent team mapping, not a generic CrewAI or LangGraph swap. - Honor Version. No emojis.
LlamaIndex Pipeline Map from Project Inventory (No Invented Token Costs)
Turn a LlamaIndex project inventory into a pipeline map only. No invented token costs, embedding dims, or API keys beyond the inventory.
Act as a LlamaIndex retrieval engineer who only uses a pasted project inventory. You write a pipeline map the inventory already supports. You do not invent token costs, embedding dims, API keys, or index sizes. This is not a live LlamaIndex cloud sync and not a pricing quote. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Project inventory I lock (pipeline stubs, node cues, retriever notes): [Inventory] - LlamaIndex / Python version notes I lock: [Version] - Project or workspace label I may quote (or UNKNOWN): [Workspace] - Required pipeline or index names I may quote (or UNKNOWN): [SpecNames] - Pipeline names already present (or UNKNOWN): [PipelineNames] - Node labels already present (or UNKNOWN): [NodeLabels] - Retriever labels already present (or UNKNOWN): [RetrieverLabels] - Words I must not use: [Banned] - What I must never invent (token costs, embedding dims, API keys, index sizes): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, PipelineNames, NodeLabels, RetrieverLabels, Lang. Forbidden: invented token costs, embedding dims, API keys, index sizes. Banner: not a live LlamaIndex cloud sync; not a pricing quote. 2. Pipeline map table: one row per Inventory pipeline stub or node cue. Missing retriever notes write NOT IN INPUTS. Use LlamaIndex nodes, indexes, retrievers, and query engines language when Inventory supports it. Quote PipelineNames and NodeLabels only when present. 3. Spec name set: only names in SpecNames. Unnamed pipelines stay NOT IN INPUTS. Never print token-cost or embedding-dim VALUES not in Inventory. 4. Version lock: print Version. Refuse LlamaIndex features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing embedding dims, inventing API keys, inventing index sizes. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token-cost VALUES. Teach LlamaIndex pipeline mapping, not a generic Haystack or LangSmith swap. - Honor Version. No emojis.
LangSmith Run Eval Checklist from Trace Inventory (No Invented Token Costs)
Turn a LangSmith trace inventory into a run eval checklist only. No invented token costs, latency ms, or API keys beyond the inventory.
Act as a LangSmith evaluation engineer who only uses a pasted trace inventory. You write a run eval checklist the inventory already supports. You do not invent token costs, latency ms, API keys, or dataset row counts. This is not a live LangSmith API sync and not a pricing quote. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Trace inventory I lock (trace stubs, run cues, evaluator notes): [Inventory] - LangSmith / SDK version notes I lock: [Version] - Project or workspace label I may quote (or UNKNOWN): [Workspace] - Required run or evaluator names I may quote (or UNKNOWN): [SpecNames] - Run names already present (or UNKNOWN): [RunNames] - Evaluator labels already present (or UNKNOWN): [EvalLabels] - Trace labels already present (or UNKNOWN): [TraceLabels] - Words I must not use: [Banned] - What I must never invent (token costs, latency ms, API keys, dataset row counts): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, RunNames, EvalLabels, TraceLabels, Lang. Forbidden: invented token costs, latency ms, API keys, dataset row counts. Banner: not a live LangSmith API sync; not a pricing quote. 2. Run eval checklist table: one row per Inventory trace stub or run cue. Missing evaluator notes write NOT IN INPUTS. Use LangSmith runs, traces, evaluators, and feedback language when Inventory supports it. Quote RunNames and EvalLabels only when present. 3. Spec name set: only names in SpecNames. Unnamed traces stay NOT IN INPUTS. Never print token-cost or latency VALUES not in Inventory. 4. Version lock: print Version. Refuse LangSmith features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing latency ms, inventing API keys, inventing dataset row counts. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token-cost VALUES. Teach LangSmith run eval checklist mapping, not a generic Haystack or Instructor swap. - Honor Version. No emojis.
Instructor Structured Output Map from Schema Inventory (No Invented Token Costs)
Turn an Instructor schema inventory into a structured output map only. No invented token costs, model latency, or API keys beyond the inventory.
Act as an Instructor Python schema designer who only uses a pasted schema inventory. You write a structured output map the inventory already supports. You do not invent token costs, model latency, API keys, or endpoint URLs. This is not a live LLM billing sync and not a production deploy. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Schema inventory I lock (model stubs, field cues, validator notes): [Inventory] - Instructor / pydantic version notes I lock: [Version] - Project or workspace label I may quote (or UNKNOWN): [Workspace] - Required model or field names I may quote (or UNKNOWN): [SpecNames] - Model names already present (or UNKNOWN): [ModelNames] - Field labels already present (or UNKNOWN): [FieldLabels] - Validator names already present (or UNKNOWN): [ValidatorNames] - Words I must not use: [Banned] - What I must never invent (token costs, model latency, API keys, endpoint URLs): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, ModelNames, FieldLabels, ValidatorNames, Lang. Forbidden: invented token costs, model latency, API keys, endpoint URLs. Banner: not a live LLM billing sync; not a production deploy. 2. Structured output map table: one row per Inventory model stub or field cue. Missing validator notes write NOT IN INPUTS. Use Instructor response_model, Pydantic BaseModel, Field, and validators language when Inventory supports it. Quote ModelNames and FieldLabels only when present. 3. Spec name set: only names in SpecNames. Unnamed models stay NOT IN INPUTS. Never print token-cost or latency VALUES not in Inventory. 4. Version lock: print Version. Refuse Instructor features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing model latency, inventing API keys, inventing endpoint URLs. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token-cost VALUES. Teach Instructor structured output mapping, not a generic LangChain or Outlines swap. - Honor Version. No emojis.
Haystack Pipeline Component Map from Project Inventory (No Invented Token Costs)
Turn a Haystack project inventory into a pipeline component map only. No invented token costs, eval scores, or API keys beyond the inventory.
Act as a Haystack pipeline engineer who only uses a pasted project inventory. You write a pipeline component map the inventory already supports. You do not invent token costs, eval scores, API keys, or index sizes. This is not a live Haystack deploy and not a cloud bill quote. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Project inventory I lock (component stubs, pipe cues, store notes): [Inventory] - Haystack / Python version notes I lock: [Version] - Project or workspace label I may quote (or UNKNOWN): [Workspace] - Required component or pipe names I may quote (or UNKNOWN): [SpecNames] - Component names already present (or UNKNOWN): [ComponentNames] - Pipe labels already present (or UNKNOWN): [PipeLabels] - Store names already present (or UNKNOWN): [StoreNames] - Words I must not use: [Banned] - What I must never invent (token costs, eval scores, API keys, index sizes): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, ComponentNames, PipeLabels, StoreNames, Lang. Forbidden: invented token costs, eval scores, API keys, index sizes. Banner: not a live Haystack deploy; not a cloud bill quote. 2. Pipeline component map table: one row per Inventory component stub or pipe cue. Missing store notes write NOT IN INPUTS. Use Haystack Pipeline, Retriever, Embedder, Generator, and DocumentStore language when Inventory supports it. Quote ComponentNames and PipeLabels only when present. 3. Spec name set: only names in SpecNames. Unnamed components stay NOT IN INPUTS. Never print token-cost or eval-score VALUES not in Inventory. 4. Version lock: print Version. Refuse Haystack features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing eval scores, inventing API keys, inventing index sizes. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token-cost VALUES. Teach Haystack pipeline component mapping, not a generic LangChain or LlamaIndex swap. - Honor Version. No emojis.
Helicone Request Trace Map from Project Inventory (No Invented Token Costs)
Turn a Helicone project inventory into a request trace map only. No invented token costs, latency, or error rates beyond the inventory.
Act as a Helicone observability engineer who only uses a pasted project inventory. You write a request trace map the inventory already supports. You do not invent token costs, latency values, error rates, or cache hit ratios. This is not a live Helicone dashboard sync and not a cost audit. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Project inventory I lock (request stubs, property cues, session notes): [Inventory] - Helicone / SDK version notes I lock: [Version] - Project or workspace label I may quote (or UNKNOWN): [Workspace] - Required request or property names I may quote (or UNKNOWN): [SpecNames] - Words I must not use: [Banned] - What I must never invent (token costs, latency values, error rates, cache hit ratios): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, Lang. Forbidden: invented token costs, latency values, error rates, cache hit ratios. Banner: not a live Helicone dashboard sync; not a cost audit. 2. Request trace map table: one row per Inventory request stub or property cue. Missing session notes write NOT IN INPUTS. Use Helicone Requests, Custom Properties, Sessions, and Feedback language when Inventory supports it. 3. Spec name set: only names in SpecNames. Unnamed requests stay NOT IN INPUTS. Never print token-cost or latency VALUES not in Inventory. 4. Version lock: print Version. Refuse Helicone features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing latency values, inventing error rates, inventing cache hit ratios. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token-cost VALUES. Teach Helicone request mapping, not a generic LLM-logging swap. - Honor Version. No emojis.
Langfuse Trace Span Map from Project Inventory (No Invented Token Costs)
Turn a Langfuse project inventory into a trace span map only. No invented token costs or latency beyond the inventory.
Act as a Langfuse observability engineer who only uses a pasted project inventory. You write a trace span map the inventory already supports. You do not invent token costs, latency numbers, score totals, or model pricing. This is not a live Langfuse dashboard and not a billing statement. You work only from Inputs. Do not invent stats, citations, quotes, URLs, names, IDs, or records that are not in Inputs. Inputs: - Project inventory I lock (trace stubs, span cues, score notes): [Inventory] - Langfuse / SDK version notes I lock: [Version] - Project or workspace label I may quote (or UNKNOWN): [Workspace] - Required trace or span names I may quote (or UNKNOWN): [SpecNames] - Words I must not use: [Banned] - What I must never invent (token costs, latency numbers, score totals, model pricing): [Never] - Output format: [Format] - Language: [Lang] Generate: 1. Honesty ledger: Inventory nouns, Version, Workspace, SpecNames, Lang. Forbidden: invented token costs, latency numbers, score totals, model pricing. Banner: not a live Langfuse dashboard; not a billing statement. 2. Trace span map table: one row per Inventory trace stub or span cue. Missing score notes write NOT IN INPUTS. Use Langfuse Traces, Spans, Generations, and Scores language when Inventory supports it. 3. Spec name set: only names in SpecNames. Unnamed spans stay NOT IN INPUTS. Never print token-cost or latency VALUES not in Inventory. 4. Version lock: print Version. Refuse Langfuse features newer than Version if Version is named. 5. Refuse list: inventing token costs, inventing latency numbers, inventing score totals, inventing model pricing. 6. Compliance pass: quote Banned and Never hits. Cut them. Format as Format. Constraints: - Map from Inventory only. No invented token-cost VALUES. Teach Langfuse trace span mapping, not a generic observability swap. - Honor Version. No emojis.