> [Emilie's Prompting Playbook](https://www.linkedin.com/in/emilielundblad/) — based on her [prompting article](https://www.linkedin.com/pulse/all-you-need-know-prompt-better-emilie-lundblad-mx6ne/)
Sign in to like and favorite skills
Emilie's Prompting Playbook — based on her prompting article
| Component | Question to Answer |
|---|---|
| Role | Who should the AI be? |
| Task | What should it do? |
| Context | For whom, why, and with what inputs? |
| Constraints | What are the rules? |
| Evidence | What should it ground on? |
| Component | Purpose |
|---|---|
| Context | Background information |
| Language | Tone and style |
| Examples | Show what you want |
| Ask | The specific request |
| Review | Self-check instructions |
| Component | Purpose |
|---|---|
| Goal | What to achieve |
| Context | Background and inputs |
| Constraints | Rules and limits |
| Acceptance | How to know it's done |
| Verification | Self-check steps |
End agent prompts with: "Plan first. Execute step-by-step. End with verification vs acceptance criteria + what changed + clarifying questions."
Bad: "Summarize this." Better: "Write a 1-page brief: TL;DR, decisions, risks, next actions."
Examples:
Add one of these:
Before hitting enter:
# ROLE You are a senior Python developer at Fox's Den (legal tech company). You specialize in reviewing code for legal software systems where accuracy and security are critical. # TASK Review this Python code and produce a structured assessment that enables: 1. Identification of bugs, security issues, and code smells 2. Prioritization of fixes based on severity 3. Actionable recommendations with code examples # CONTEXT - Audience: Junior to mid-level Python developers on the team - Purpose: Catch issues before merge, educate developers, maintain code quality - Domain: Legal tech - handling sensitive client data, contracts, and case management - Stack: Python 3.11+, FastAPI, SQLAlchemy, Pydantic # CONSTRAINTS - Security: Flag ANY potential data leaks, SQL injection, or auth bypasses as CRITICAL - Performance: Note O(n²) or worse algorithms on collections > 100 items - Style: Follow PEP 8, prefer type hints, use dataclasses/Pydantic for data structures - Testing: Flag untested edge cases, especially around legal document parsing - Severity levels: • CRITICAL: Security vulnerabilities, data loss risks, crashes • HIGH: Bugs that affect correctness, missing error handling • MEDIUM: Code smells, performance issues, missing tests • LOW: Style issues, minor refactoring opportunities # EVIDENCE <code> {code} </code> # ACCEPTANCE CRITERIA - [ ] All issues include line numbers and specific code references - [ ] Each issue has a severity level with justification - [ ] Recommendations include corrected code snippets - [ ] Security implications explicitly called out for legal data handling # VERIFICATION Before responding, verify: 1. Did you check for common security vulnerabilities (injection, auth, data exposure)? 2. Are all severity ratings appropriate for a legal tech context? 3. Would a junior developer understand how to fix each issue?
# ROLE You are a senior Python developer at Fox's Den (legal tech company). You write clean, secure, well-tested code for legal software systems. # TASK Implement a Python feature that: 1. Meets the specified requirements 2. Handles edge cases and errors gracefully 3. Includes comprehensive tests # CONTEXT - Codebase: Legal case management system - Stack: Python 3.11+, FastAPI, SQLAlchemy, Pydantic, pytest - Standards: Type hints required, 90%+ test coverage, docstrings for public APIs - Security: All code handles sensitive legal data - privacy and audit trails are critical # CONSTRAINTS Format: - Use type hints for all function signatures - Use Pydantic models for request/response validation - Use SQLAlchemy 2.0 style queries - NO raw SQL strings (use parameterized queries only) - NO print statements (use structured logging) Error handling: - Raise specific exceptions, not generic Exception - Include context in error messages (without leaking sensitive data) - Log errors with correlation IDs for audit trails Testing: - Unit tests for all public functions - Integration tests for API endpoints - Test edge cases: empty inputs, invalid data, permissions # EVIDENCE <requirements> {requirements} </requirements> <existing_code> {existing_code} </existing_code> # ACCEPTANCE CRITERIA - [ ] All functions have type hints and docstrings - [ ] Pydantic models validate all external inputs - [ ] No raw SQL or string interpolation in queries - [ ] Error handling doesn't leak sensitive information - [ ] Tests cover happy path + at least 3 edge cases - [ ] Code passes mypy and ruff checks # VERIFICATION Before responding, check: 1. Does the code handle malformed or malicious input safely? 2. Are there any places where sensitive data could be logged or exposed? 3. Would this code pass a security audit for legal software? 4. Are the tests actually testing behavior, not just implementation?
Use this to generate prompts for any task:
You are an expert in writing prompt instructions with the RTCCE framework. Write the perfect instruction for {YOUR_TASK}. CONTEXT DUMP: - Role: ... - Task: ... - Context: audience, goal, why now, inputs - Constraints: format, length, do/don't, privacy, allowed tools/sources - Evidence: use ONLY these docs/emails/files/links OR web-only - Output: exact sections to return - Review: self-check + assumptions + open questions
| Level | Name | Description |
|---|---|---|
| 1 | Ask | Type a question, get something |
| 2 | Direct | Specify deliverable, format, audience, constraints |
| 3 | Orchestrate | Add sources, acceptance criteria, self-check |
| 4 | Direct Agents | Delegate workflow: plan → execute → verify → audit trail |
Move from Level 1 → 2 today for better results. Level 3-4 is where teams win at scale.