Lab 07: Agentic SDLC

Use agents as engineering leverage without losing ownership.

1. Write a problem statement for a small feature.
2. Ask an agent to inspect the codebase and return a map.
3. Review the map manually.
4. Ask an agent to propose an implementation plan.
5. Reject or edit weak parts of the plan.
6. Implement a bounded slice yourself or with an agent.
7. Ask an agent for a code review.
8. Run tests and runtime verification.
9. Write what the agent got wrong.

Use the agent prompt pack as a starting point.

Use one explorer prompt:

Inspect this repo for the ticket creation flow. Do not edit files. Return entry
points, data flow, validation points, tests, run commands, and risks.

Use one reviewer prompt:

Review this diff for bugs, missing tests, security issues, and behavior
regressions. Findings first. Reference file paths and lines.

Create agentic-sdlc-log.md with:

• Goal.
• Context given.
• Agent outputs.
• Human review.
• Accepted changes.
• Rejected changes.
• Verification commands.
• Remaining risks.

The final feature is verified and you can explain exactly how agents helped and where they were wrong or incomplete.