Claude Prompt-Fix Workflow

$750-1500 USD

Posted 1 minute ago

•

Ends in 6 days

$750-1500 USD

Paid on delivery

I use Higgsfield every day to plan shots, yet Claude has become unreliable: it ignores even the simplest “one-sentence only” request and seems to have lost all logic when retrieving or summarising information. I need a clear, repeatable workflow that restores discipline so Claude consistently follows my prompts word-for-word. Your task is to design and document that workflow—from the exact wording of the system prompt through any step-by-step guardrails or fallback checks that keep it on track. Please include tested examples tailored to film work (e.g., describing a single shot, pulling data from my notes, obeying camera directions) so I can drop them straight into production. Deliverables • A concise playbook outlining each stage of the workflow • At least three proven prompt templates (single-sentence answer, file-lookup, multi-step follow-through) • Troubleshooting tips for when Claude drifts or hallucinate I’ll judge success by running your workflow against fresh queries; if Claude stays within the requested sentence length or step list without inventing extras, the job is done.

Project ID: 40818535

About the project

25 proposals

Open for bidding

Remote project

Active 5 secs ago

Place your bid

Bid amount Email address

Benefits of bidding on Freelancer

Set your budget and timeframe

It's free to sign up and bid on jobs

25 freelancers are bidding on average $912 USD for this job

User Avatar
Flag of BANGLADESH

AI workflow automation and chatbot development are central to my work, making me well suited to turn Claude’s inconsistent responses into a disciplined, repeatable production workflow for your Higgsfield film-planning process. I can create a concise playbook covering the system prompt, response constraints, validation guardrails, and fallback checks without adding unnecessary complexity. My AI content creation experience supports practical prompt templates for single-sentence shot descriptions, structured note and file lookups, and multi-step follow-through. I will tailor the examples to film work, including camera directions, shot details, strict sentence limits, source-grounded summaries, and checks that flag missing information rather than allowing invented content. I would be glad to deliver a clear, production-ready Claude prompt-fix workflow that you can test against fresh queries and reuse consistently.

User Avatar
Flag of UKRAINE

Hello, Claude’s failure to honor constraints like “one sentence only” usually requires more than a stronger prompt, it needs explicit output contracts, staged retrieval, and validation before delivery. I’ll create a concise production playbook covering the system prompt, instruction hierarchy, context handling, fallback behavior, and checks for sentence counts, step limits, and unsupported details. The templates will be tailored to your Higgsfield film workflow: single-shot descriptions, note/file lookup, camera-direction compliance, and multi-step follow-through. I’ll also include fresh test cases and troubleshooting guidance for drift, hallucinations, and incomplete retrieval, without claiming results that have not been verified. I can deliver the first complete, production-ready playbook with templates and validation examples within 5 days. Best regards, Bohdan

User Avatar
Flag of INDIA

Hi, this is my daily work: I head a creative department and use Claude every day for film and ad production (shot lists, storyboard frame prompts, character and camera direction for AI video), and I build software on Claude too. Most "Claude ignores me" problems come from a few fixable causes: long chats where early rules get buried, instructions phrased as preferences instead of hard formats, notes pasted without structure so retrieval drifts, and no check step before the answer. What I'll deliver: 1. Playbook (short, practical): where to put rules (a Claude Project's custom instructions vs the message), how to phrase hard constraints so they stick, when to start a fresh chat, and how to structure your shot notes so lookups are exact. 2. System prompt for your film workflow: role, output contract (length, format, no extras), camera and shot vocabulary, and a "quote from my notes or say not found" rule to stop invented details. 3. Three tested templates, plus a couple of bonus ones: a. Single sentence answer (e.g. describe one shot) with a hard one sentence contract. b. File lookup: answers only from your notes, cites the scene or shot number, says "not in notes" instead of guessing. c. Multi step follow through: numbered steps, Claude confirms each one is done, no extra steps added. 4. Drift and hallucination troubleshooting: a short checklist and copy paste "reset" prompts for when it wanders. 5. Test sheet: 20 fresh queries run against each template with the results, so you can see it hold before you use it. Templates will be ready to drop straight into Higgsfield shot planning. Timeline: 5 days. Do you use Claude in the app (with Projects) or through the API, and can you share 2 or 3 recent prompts where it went wrong?

User Avatar
Flag of GERMANY

Claude's inconsistent obedience to output limits and note-based facts is the main difficulty here. I can create a concise playbook with a system prompt, three templates for one-sentence shots, file lookup, and multi-step follow-through, plus film examples for camera directions and fresh-query testing. We built an AI gateway with model access, usage controls, and automatic top-ups, so we are familiar with making model workflows predictable and testable. Can you provide the Claude setup and sample notes you want this workflow tested against?

User Avatar
Flag of PAKISTAN

Hi, I understand you need a dependable Claude workflow that follows strict instructions, retrieves notes correctly, and avoids adding unsupported details. I will Design the system prompt, create clear guardrails, and build fallback checks for sentence limits, step counts, file lookup, and camera directions. The playbook will include three ready-to-use film prompts, expected outputs, failure examples, and practical Troubleshooting steps. I will also structure the Automation so each response can be checked before it reaches production, while keeping the Chatbot behavior simple to reuse with Higgsfield and Claude. Are your film notes supplied to Claude as uploaded files, project knowledge, or pasted text, and which format do you use most often? Please start the chat and let’s discuss your current prompts and notes setup.

User Avatar
Flag of INDONESIA

Claude ignoring a one-sentence limit while you plan shots in Higgsfield is fixable, and I can set this workflow up for you. I can take this on for you from the first system prompt to the last troubleshooting note. I'll stay with it until Claude holds to your prompts on fresh queries. Are you using Claude in the chat app, in Projects, or through the API? Happy to go through the details with you in chat. Sela

User Avatar
Flag of BANGLADESH

AI chatbot development and AI content creation give me the right foundation to build a disciplined Claude workflow that prioritizes instruction-following, reliable retrieval, and controlled outputs. My work in AI automation also aligns with your need for a repeatable system rather than isolated prompt advice. I can create a concise playbook covering the system prompt, output constraints, validation guardrails, fallback checks, and troubleshooting for drift or hallucinations. I will tailor the examples to film production, including one-sentence shot descriptions, structured note lookups, and camera-direction prompts, while providing at least three reusable templates for single-sentence answers, file retrieval, and multi-step follow-through. As part of 360 Elevate’s broader digital and AI automation capabilities, I work with structured workflows designed to make AI systems more dependable and practical in daily operations. Choose me to turn your Claude process into clear, production-ready documentation you can apply directly in Higgsfield.

User Avatar
Flag of GERMANY

Hello, "Claude prompt-fix workflow" - the goal is predictable output, not another long system prompt that Claude can still ignore. I'll build the workflow around explicit output contracts, source-grounded retrieval and separate checks for each task. I'll write the system prompt, then create three film-focused templates for one-sentence shot descriptions, searching production notes and following multi-step camera instructions. I'll also add fallback rules for missing source material, conflicting directions and unsupported details, plus test cases that check sentence count, step count and factual accuracy. One important safeguard: retrieval answers should distinguish what is actually in the notes from what Claude is inferring. Do you want the file-lookup template to work with uploaded production notes, a connected knowledge base, or both? Looking forward to working with you. Beate

User Avatar
Flag of SERBIA

HELLO, "Claude Prompt-Fix Workflow" - Claude needs to follow exact instructions, retrieve information accurately, and stop adding unwanted explanations or invented details. I'll build a repeatable workflow covering the system prompt, instruction hierarchy, output constraints, file-retrieval rules, and fallback checks, then document it in a concise playbook with at least three ready-to-use templates. I'll tailor the examples to your film workflow in Higgsfield, including single-shot descriptions, camera directions, and retrieving information from production notes. One important safeguard is testing conflicting instructions and missing source information so Claude knows when to stop rather than guess. I'll also include troubleshooting steps for identifying and correcting prompt drift. Should the file-lookup template work only with supplied production notes, or also with external reference material? I appreciate your consideration and look forward to delivering a workflow you can test directly in production. THANKS

User Avatar
Flag of UNITED STATES

Hi, I like this project because I realized what you want in this and have already found some technical solutions for it. Reliable AI workflows are built through precise instruction design and disciplined validation, and AI development is a core part of my work. With 20+ years taking complex products from vague ideas to reliable production systems, I can turn Claude’s inconsistent behavior into a clear, repeatable workflow tailored to your Higgsfield film-planning process. I will document the complete playbook, including the system prompt, enforcement guardrails, output-format checks, and fallback handling for retrieval and summarization. The prompt templates will cover concise single-shot descriptions, grounded file lookups, and multi-step follow-through, with film-specific examples for camera directions and shot planning. I will also include practical troubleshooting for drift, unsupported claims, and invented information so the workflow is usable against fresh queries rather than merely theoretical. My focus is dependable AI integration: clear boundaries, predictable outputs, and documentation you can apply directly in production. Choose me to create a concise Claude prompt-fix workflow that restores control and makes each response follow the requested format. Best, Shayne

User Avatar
Flag of MOROCCO

Hi Adam, I’ll deliver a concise playbook, three tested Claude prompt templates (single‑sentence, file‑lookup, multi‑step) and troubleshooting tips, all tailored to your film shot notes. I’ll complete this within two weeks for $1,200. Ready to start now with a free sample template?

User Avatar
Flag of ARGENTINA

Hello, Claude's inconsistency needs to be addressed at the workflow level, not just by making prompts longer. I'd build a structured instruction system that separates output constraints, source-grounding rules, execution steps, and validation checks so each request has a clearly defined expected result. I'd create a reusable system prompt with explicit priority rules, strict output formats, and instructions for handling missing information. On top of that, I'd provide three production-ready templates covering single-sentence responses, file-based retrieval, and multi-step execution, tailored to Higgsfield and film-production workflows. For example, a camera-direction request should preserve the specified shot, movement, framing, and subject without introducing additional creative details. File lookups should distinguish information actually present in your notes from missing information rather than inventing answers. One important safeguard is automated output validation with retry and fallback instructions. I'll test fresh examples, including ambiguous requests and conflicting instructions, to identify where Claude drifts and refine the workflow accordingly. You'll receive a practical playbook, reusable templates, troubleshooting guidance, and clear implementation instructions. Nicolas

User Avatar
Flag of UNITED STATES

Reliable Claude behavior depends on prompt engineering backed by explicit validation, and my AI application experience includes Claude, GPT, LangChain, RAG, AI agents, and chatbot development. I can turn your Higgsfield film workflow into a concise, production-ready playbook rather than a collection of vague prompting tips. I will document a precise system prompt, reusable templates for one-sentence shot descriptions, note/file retrieval, and multi-step follow-through, plus guardrails that enforce output length, step counts, source boundaries, camera-direction fidelity, and “do not invent” rules. The examples will be tailored to filmmaking so you can test them directly against shot planning, production notes, and camera instructions. I will also include practical troubleshooting for drift, unsupported summaries, missing context, and noncompliant formatting, with fallback checks that make failures visible and repeatable. My background in AI-powered applications, prompt engineering, RAG, and AI chatbots makes me well suited to create a workflow that is clear, testable, and easy to maintain. I’m ready to build the Claude prompt-fix playbook you can use in production.

User Avatar
Flag of UNITED STATES

Hi there, I'd be happy to help move "Claude Prompt-Fix Workflow" forward. I'm Silas, a Full Stack Developer and AI Engineer. I build web applications, backend services, API integrations, and AI-powered features using technologies like React, Node.js, TypeScript, and Python. My focus isn't just on getting features working. I also pay attention to performance, reliability, code quality, and how everything fits together in the long run. Could you tell me a bit about the current stage of your project? Kind regards, Silas

User Avatar
Flag of BANGLADESH

Reliable AI automation and chatbot development is central to my work, making me well suited to build a disciplined Claude workflow that follows exact output constraints instead of adding unwanted content. I can turn your Higgsfield-to-Claude process into a concise, reusable playbook covering system instructions, validation guardrails, fallback checks, and recovery prompts. My AI Content Creation and AI Development skills support practical templates for your film-production use cases: one-sentence shot descriptions, controlled retrieval and summarisation from notes, and multi-step prompts that preserve camera directions and required order. I will also document troubleshooting for drift, unsupported file lookups, missing information, and hallucinated details, with clear rules for refusing to invent content and checking sentence or step limits. I am ready to create a production-ready prompt-fix workflow with at least three copy-and-use templates and examples tailored to shot planning, so you can test Claude consistently against fresh queries.

内容来自 Freelancer News,版权归原作者所有。

阅读原文 ↗

把喜欢的内容,留在你的新标签页。

打开我的订阅