April 22, 2026
Learn how to work effectively with AI coding agents using a disciplined, spec-driven workflow. This workshop covers specification writing, iterative prompt planning, and using agents to review and improve code quality.
Reference prompt collections for data science workflows with AI coding agents.
Build a command-line temperature anomaly detector from scratch using a disciplined, spec-first workflow. You will iteratively develop a specification through conversation, critique it, plan an implementation, and then build it with AI assistance.
Create a CLI tool that reads a CSV of daily temperature readings, detects anomalies (days more than 2 standard deviations from a 30-day rolling average), and outputs a list of flagged dates, deviation magnitudes, and an ASCII chart. The tool must validate the schema, handle missing values explicitly, and include tests.
Use these prompts in sequence with Claude or another AI coding agent.
Ask one question at a time to iteratively develop a thorough, step-by-step specification.
Ask me one question at a time so we can develop a thorough, step-by-step spec for this idea. Each question should build on my previous answers, and our end goal is to have a detailed specification I can hand off to a developer. Let's do this iteratively and dig into every relevant detail. Remember, only one question at a time.
Consolidate the conversation into a developer-ready specification document.
Now that we've wrapped up the brainstorming process, can you compile our findings into a comprehensive, developer-ready specification? Include all relevant requirements, architecture choices, data handling details, error handling strategies, and a testing plan so a developer can immediately begin implementation.
Have the agent act as a senior architect and identify gaps, ambiguities, and risks before implementation.
You are a senior software architect reviewing a developer-ready specification for an agentic coding system. Evaluate whether the specification is: Clear, Complete, Consistent, Correct, Testable, and Unambiguous. Identify missing requirements, ambiguities, inconsistencies, hidden assumptions, edge cases, and testability weaknesses. For each issue: quote the problematic section, explain the risk, and propose a precise improvement. Conclude with a readiness assessment. [Attach specification.md]
Generate a step-by-step blueprint broken into small, TDD-friendly implementation prompts.
Draft a detailed, step-by-step blueprint for building this project. Break it into small, iterative chunks that build on each other. From here, provide a series of prompts for a code-generation LLM that will implement each step in a test-driven manner. Prioritize best practices, incremental progress, and early testing. Each prompt should build on the previous prompts and end with wiring things together. [Attach specification]
Example artifacts produced by running through the exercise.
Use Claude Code as a code reviewer. A deliberately flawed codebase contains six intentional issues spanning security, error handling, test quality, and project hygiene. Your job is to prompt the agent, judge its findings, and then use it to apply fixes.
| # | Issue | Category |
|---|---|---|
| 1 | requirements.txt instead of pyproject.toml | Toolchain compliance |
| 2 | SQL query built with f-string formatting | Security (injection) |
| 3 | Bare except: pass swallowing errors | Error handling |
| 4 | Missing type hints on public functions | Code standards |
| 5 | Test that passes but asserts the wrong variable | Test quality |
| 6 | No .gitignore — venv/cache would be committed | Git hygiene |
Prompt Claude Code with:
Review this codebase against AGENTS.md for compliance. Check for: security issues, exception handling problems, test quality, dependency management, type hint coverage, and Git hygiene. For each issue, explain the risk and provide the exact file and line number.
Read the agent's findings and cross-reference with the code. Did it catch all six? Did it miss any? Did it flag anything that is actually fine (false positive)?
Ask the agent to fix the highest-severity issue: apply the fix, run tests, and commit. Alternatively, enter planning mode, review a plan to fix all issues, then execute it.
If the fix requires a dependency change, observe the approval gate in action — the agent should pause and ask before installing anything.
Alternative: use /simplify or /review in Claude Code to launch subagents for code reuse, quality, and efficiency review.
Reference materials from the initial formal process walkthrough, including the step-by-step guide and the supporting codex.