DeepSeek V4 Pro Prompt Generator

You are a senior software engineer agent with access to the file system, terminal, and test runner. Complete the following multi-step task autonomously: 1. READ all files in the /src directory and build a complete dependency map 2. IDENTIFY all instances of the deprecated `fetchData()` function (12 call sites across 8 files) 3. PLAN the migration to the new `fetchDataV2()` API — list every file change needed before touching any file 4. EXECUTE the migration one file at a time, running the test suite after each file change 5. If any test fails, diagnose the root cause and fix it before proceeding to the next file 6. PRODUCE a migration summary: files changed, tests passed/failed, any manual steps still required Thinking effort: high. Report status after each step using [STEP N COMPLETE] markers.

I am providing a complete M&A agreement (approximately 800,000 tokens). Analyse the full document and produce: 1. RISK FLAGS — every clause that creates material liability for the acquirer, with section references 2. UNUSUAL TERMS — clauses that deviate from standard M&A boilerplate, explained in plain language 3. MISSING PROTECTIONS — standard acquirer protections absent from this agreement 4. REPRESENTATIONS TABLE — all seller reps/warranties with their survival periods and caps 5. CONDITION PRECEDENTS — complete list of conditions to closing, colour-coded by status (clear/amber/red) 6. EXECUTIVE SUMMARY — a 1-page brief suitable for presenting to a non-legal board Cite specific section numbers for every finding. [Paste full agreement]

Repository: Flask web application, 15,000 lines. Bug report: 'POST /api/upload endpoint returns 500 error for files larger than 2MB but no error appears in the application logs.' Agent task: 1. Locate the upload endpoint and all middleware in the request chain 2. Identify all file size limits set in Flask config, nginx config, and any middleware (look for MAX_CONTENT_LENGTH, client_max_body_size, etc.) 3. Trace why the error is swallowed before reaching the logger 4. Write the fix: both the size limit configuration and the error handler that was missing 5. Write a regression test for this exact scenario 6. Output: a unified diff of all changed files + the new test file Thinking effort: max. Show reasoning before each investigative step.

I'm providing 25 research papers, 3 technical whitepapers, and 12 analyst reports on AI regulation in the European Union (total approximately 480,000 tokens). Synthesise this corpus: 1. TIMELINE — key regulatory milestones from the 2021 AI Act draft to current 2026 enforcement 2. CONSENSUS — findings agreed across ≥70% of sources 3. CONTRADICTIONS — where expert opinion diverges, with an explanation of what drives each position 4. ENFORCEMENT GAPS — requirements in the AI Act that lack clear enforcement mechanisms 5. COMPANY IMPLICATIONS — a matrix of required actions by company type (foundation model provider, deployer, high-risk use case) 6. OPEN QUESTIONS — the 5 most significant unresolved policy questions as of mid-2026 [Paste full corpus]

Thinking effort: low. Analyse this user feedback survey (150 responses, pasted below) and produce a structured output: - Top 3 satisfaction drivers (with frequency count) - Top 3 friction points (with frequency count) - NPS category breakdown: Promoters / Passives / Detractors - Three concrete product changes the data most directly supports Format as JSON with keys: satisfaction_drivers, friction_points, nps_breakdown, recommended_changes. [Paste survey data]

You are an integration agent working in OpenAI Responses API format. Complete this multi-step integration task: Task: Connect our CRM (HubSpot) to our analytics platform (Mixpanel) so that deal stage changes in HubSpot automatically trigger Mixpanel events. Step 1: Propose the webhook payload structure from HubSpot's deal.propertyChange event Step 2: Write the transformation function that maps HubSpot fields to Mixpanel event properties Step 3: Write the serverless function handler (Node.js) with error handling, retry logic (3 attempts with exponential backoff), and dead-letter queue fallback Step 4: Write the unit tests covering: successful event, HubSpot timeout, Mixpanel 429, and malformed payload Step 5: Write the infrastructure-as-code (Terraform) for the Lambda + SQS dead-letter queue Output each step as a separate code block with file name as the comment header.

I'm providing the complete source of a 40,000-line Node.js microservices codebase (8 services, shared libraries, Docker and Kubernetes configs). Conduct a senior architect-level review: 1. SERVICE MAP — diagram the service dependencies and data flow in text-based ASCII 2. COUPLING ISSUES — identify tightly coupled services that violate microservice independence principles 3. SINGLE POINTS OF FAILURE — every component where a crash would cascade to multiple services 4. SCALING BOTTLENECKS — which services will fail first under 10× load and why 5. SECURITY SURFACE — enumerate every external API surface, auth mechanism, and data store access pattern 6. MIGRATION PLAN — the 5 highest-ROI architectural changes, ordered by risk × impact [Paste full codebase]

Thinking effort: max. I am founding a B2B SaaS company targeting mid-market legal firms (50–500 employees) with AI-powered contract review. Available capital: $2M seed. Current team: 2 engineers, 1 founder (former M&A lawyer). Produce a complete 18-month strategic plan: 1. GO-TO-MARKET — ICP definition, acquisition channels, CAC targets, and sales motion (PLG vs direct) 2. PRODUCT ROADMAP — feature sequencing from MVP to Series A product, with rationale for each priority 3. FINANCIAL MODEL — monthly projections for ARR, burn, and headcount needed to reach $2M ARR 4. COMPETITIVE MOAT — how to defend against Harvey AI, Ironclad, and a well-funded new entrant 5. RISK REGISTER — top 5 existential risks and mitigation strategies 6. SERIES A NARRATIVE — the story a top-tier VC would find compelling 18 months from now Show your reasoning for all major assumptions.

I'm providing a Python data science library (approximately 120,000 tokens of source code) that has no documentation. Generate complete documentation: 1. MODULE OVERVIEW — one paragraph per module explaining its purpose and public API 2. FUNCTION REFERENCE — for every public function: description, parameters (type + purpose), return value, exceptions raised, one usage example 3. QUICKSTART GUIDE — the 5 most common use cases, each as a complete runnable code snippet 4. ARCHITECTURE DIAGRAM — text-based diagram showing how modules depend on each other 5. MIGRATION NOTES — anything that appears to be a breaking change vs the previous version (infer from comments and deprecated markers) Format as reStructuredText ready for Sphinx. [Paste source code]

You are a code review agent. I'm providing a pull request diff (approximately 3,000 lines across 15 files). Complete a thorough review: 1. CORRECTNESS — identify any logic errors, off-by-one errors, or incorrect assumptions 2. SECURITY — flag any injection vulnerabilities, missing input validation, or insecure defaults introduced by this PR 3. PERFORMANCE — identify N+1 queries, missing indexes, or O(n²) algorithms where O(n log n) is achievable 4. TEST COVERAGE — which new code paths have no corresponding test; write the missing test cases 5. STYLE — list style violations (inconsistent naming, unused imports, etc.) as a separate lower-priority list 6. VERDICT — APPROVE, REQUEST CHANGES, or NEEDS DISCUSSION, with a one-sentence rationale Thinking effort: high. Output in GitHub PR review comment format. [Paste diff]

I'm providing three competitor annual reports (total approximately 350,000 tokens). Extract competitive intelligence: 1. REVENUE BREAKDOWN — product line revenue, growth rates, and margin by segment for each competitor 2. R&D FOCUS — what each competitor is investing in (infer from capex, hiring disclosures, product mentions) 3. CUSTOMER CONCENTRATION — any customer accounting for >10% of revenue; retention language 4. STRATEGIC SHIFTS — changes in language between last year's report and this year's (signals of pivot) 5. WEAKNESSES REVEALED — what each competitor's own risk factors reveal about their vulnerabilities 6. OPPORTUNITY GAPS — markets or customer segments competitors explicitly acknowledge they're not serving [Paste three reports]

Thinking effort: high. I have a system design interview for a Staff Engineer role at a large-scale fintech company in 4 days. The role involves designing payment processing infrastructure handling 50,000 transactions per second. Prepare me comprehensively: 1. CORE CONCEPTS — the 8 most critical distributed systems concepts I must be fluent in for this specific domain 2. PRACTICE QUESTIONS — 5 realistic system design questions with full example answers showing how to structure the 45-minute interview 3. COMMON MISTAKES — the top 5 mistakes candidates make in payment system design interviews and how to avoid each 4. DEPTH SIGNALS — the specific technical details that signal seniority in this domain (not generic advice) 5. CLARIFYING QUESTIONS — the questions I should ask the interviewer at the start, with the reasoning behind each Tailor all advice to payment systems specifically, not generic system design.

You are a terminal agent. Complete the following shell task autonomously, showing each command before executing it: Task: I have 50,000 log files in /var/log/app/ each named app-YYYY-MM-DD-HH.log. Some are gzipped (.gz extension), some are plain text. Find all log entries containing 'ERROR' or 'CRITICAL' from the last 7 days, deduplicate identical error messages, count occurrences of each unique error, and produce a sorted report (most frequent first) saved to /tmp/error-report-$(date +%Y%m%d).txt Constraints: memory limit 512MB (don't load all files at once), must work on both Linux and macOS, no Python allowed. Produce the complete shell script with inline comments, then run it.

Natural language spec: Build a command-line expense tracker in Python. Users can: add expenses (amount, category, description, date), list expenses filtered by date range or category, see a monthly summary with category breakdown, export to CSV, and set a monthly budget with a warning when over 80% spent. Deliver: 1. Complete working Python application (single file, no external dependencies except stdlib) 2. Unit test file with ≥90% coverage of all functions 3. README with installation, usage examples for every feature, and example output screenshots (as ASCII) Thinking effort: high. Write clean, documented, production-quality Python. Start with a design plan before writing any code.

A data pipeline that transforms sales data from 3 source systems (Salesforce, SAP, and a custom PostgreSQL DB) into a unified analytics warehouse has started producing incorrect revenue figures since August 1, 2026. The discrepancy is 3.7% higher than expected. Agent investigation task: 1. Map all data transformation steps from source to warehouse 2. Identify which transformation introduces the 3.7% discrepancy (provide a hypothesis for each step) 3. Write SQL queries to validate or refute each hypothesis 4. Once the root cause is identified, write the fix and a data quality check that would have caught this earlier 5. Write a post-mortem document: what happened, why it wasn't caught, what prevents recurrence [Paste pipeline code and schema definitions]

I'm providing a 400-page English-language product manual for industrial equipment (approximately 200,000 tokens). Translate it to Brazilian Portuguese with full cultural and technical adaptation: 1. TRANSLATION — complete translation maintaining technical accuracy (not literal word-for-word) 2. CULTURAL ADAPTATION — identify any measurements (imperial → metric), electrical standards (110V contexts → 220V), and safety regulatory references that need localisation for Brazil 3. GLOSSARY — create a bilingual technical glossary of all domain-specific terms used 4. CONSISTENCY CHECK — flag any terms used inconsistently in the original English that should be standardised before translation 5. REVIEWER NOTES — any passages where the English source is ambiguous and the translation required an interpretive choice [Paste full manual]

Thinking effort: max. Analyse the following AI ethics dilemma with maximum rigor: Scenario: A hospital AI triage system has been shown to be 12% less accurate for patients with darker skin tones due to training data bias. Replacing it immediately with an unbiased system will cost $4M and take 8 months. During those 8 months, the current system will incorrectly triage an estimated 240 patients, of whom 4–6 may experience serious harm. Delaying replacement costs lives; replacing immediately disrupts operations and may also cause harm through the transition period. Analyse through 5 distinct ethical frameworks (utilitarian, Kantian, virtue ethics, care ethics, contractualism), identify where frameworks converge and diverge, present the strongest argument for each action, identify hidden assumptions in the framing, and conclude with a decision recommendation with explicit reasoning about which framework considerations should take precedence and why.

I'm providing a FastAPI Python application (approximately 8,000 lines, 45 endpoints). Generate complete OpenAPI 3.1 documentation: 1. INFO BLOCK — title, version, description, contact, license 2. PATHS — for every endpoint: summary, description, parameters (query/path/header/body) with types and validation rules, all possible response codes with schema examples 3. SCHEMAS — complete JSON Schema definitions for all request and response models 4. SECURITY SCHEMES — document all auth mechanisms detected in the code 5. EXAMPLES — for every POST/PUT endpoint, provide a realistic request body example and the expected successful response Output as valid YAML. Infer all documentation from the code — do not invent behaviour not present in the implementation. [Paste application code]

I'm providing 180 job postings from AI companies scraped August 2026 (total approximately 280,000 tokens). Extract structured data: For each posting output: - company_name - role_title - seniority_level (IC1–IC7 or manager equivalent) - required_skills (array, normalised) - preferred_skills (array, normalised) - location (city or remote) - estimated_salary_range (if disclosed) - mentions_ai_tools (boolean + which tools) After processing all postings: 1. Top 10 most-demanded skills across the corpus 2. Salary distribution by seniority level 3. Remote vs on-site ratio by company type 4. Most common AI tools mentioned as requirements Output the per-posting data as JSON Lines, the aggregate analysis as structured markdown. [Paste job postings]

You are a dependency upgrade agent. Repository: Python 3.10 project with 47 direct dependencies in requirements.txt. Complete this upgrade cycle autonomously: 1. READ requirements.txt and identify all dependencies with available upgrades 2. CLASSIFY each upgrade as: patch (safe), minor (test first), major (breaking change risk) 3. For each major upgrade, READ the changelog and summarise breaking changes relevant to this codebase 4. EXECUTE patch upgrades immediately (no review needed) 5. For minor upgrades: upgrade, run tests, revert if tests fail, log outcome 6. For major upgrades: propose specific code changes needed before upgrading 7. PRODUCE a final report: what was upgraded (with old → new versions), what was skipped and why, what needs manual review Thinking effort: high. Begin with patch upgrades, work upward by risk level.