Agentic Coding

Sample Repository Specification

This document specifies what the practical session project should contain for each day's exercises. It is a specification, not a built repository. Actual repository construction is a separate future effort.

The course is deliverable without a purpose-built sample repository. Every exercise includes fallback options for attendees who bring their own projects or use open source codebases.

Requirements by Day

Day Exercise What's Needed Why Fallback
1 Installation and First Contact 5–10 source files + test suite (any language) Agent needs enough code to demonstrate multi-file awareness Any personal project with tests
1 Assistive vs Agentic Comparison Same project + access to a second AI tool Side-by-side comparison of assistive and agentic modes Facilitator demonstrates the comparison
2 Write CLAUDE.md From Scratch Same project, with any existing CLAUDE.md removed Attendees need to experience the before/after of adding context Rename existing CLAUDE.md temporarily
2 Cross-Project Contamination A second project directory (different domain or language) Contamination requires switching between distinct projects Facilitator demonstrates
3 CLAUDE.md Audit The CLAUDE.md created in Day 2 Exercise 1 Audit requires a real file with both style and architecture rules Provided sample CLAUDE.md at different lifecycle stages
3 Highlander Analysis 10–15 source files with potential duplication Agent needs enough code to find duplicated logic Intentionally create a duplicate function, then ask the agent to find it
3 Write Your First Skill A repeated task in the project Skill writing requires a real procedure to codify Use a generic task (run tests with specific flags, create a new module from template)
4 Define and Execute Steel Thread Same project with a bounded task to implement Steel thread exercise requires a real implementation target Add a new API endpoint, write tests for an untested module, extract duplicated logic
4 Session Lifecycle Ritual Same project (continuation from Exercise 1) Restart prompt requires a real session to end and resume Works with any project
4 Evaluation Framework Self-Assessment Honest view of the current setup (CLAUDE.md, skills, last week's sessions) Scoring requires real practice to assess against the framework's dimensions Works for any attendee; the optional stagnation stretch (ACI-009) uses git history — eg the Intent project's (28 edits over 10 months)
5 Scaffold a Two-Node Whiteboard A scratch git repo you can create directories and files in The whiteboard lives in a directory; the exercise builds it from scratch Any empty or scratch repository
5 Claim, Ask, Announce The same scratch whiteboard (optionally two terminals) Running the coordination verbs live; single-writer removes contention Simulate both nodes by hand in one terminal
5 Provoke and Prevent a Collision The same scratch whiteboard Seeing the claim check stop a live Highlander violation Walk the collision through by hand
5 Independent Verification Pass A recently "done" work product (the Day 4 steel thread, or a sample) Verifying requires real as-built code to read at file:line Use a provided sample diff with a planted discrepancy