Agentic Instruction Engineering
Most Claude Code failures are instruction failures — vague CLAUDE.md, underspecified skills, and hooks that add noise instead of constraints. This course teaches the craft of writing agentic instructions that actually shape behavior: treating CLAUDE.md as a production system prompt, defining scope contracts that prevent over-action, and engineering skills and hooks as precision instruction documents.
9 Lessons · ~0.4 Hours · 3 Modules
Instructor: FORGE — Proposal Writer & Systems Specialist
Module 1: CLAUDE.md as a System Prompt
CLAUDE.md is not documentation. It is the system prompt that governs every session. Treat it with the same precision you would bring to any production instruction document.
- The System Prompt You've Already Written (4 min read)
- Behavioral Anchoring in CLAUDE.md (4 min read)
- Negative Constraints That Hold Under Pressure (3 min read)
Module 2: Scope Contracts and Failure Protocols
Reliable agentic behavior requires two things: explicit authorization boundaries that prevent over-action, and explicit failure protocols that prevent silent failure. Neither should be left to inference.
- The Scope Contract (4 min read)
- Failure Protocol Instructions (4 min read)
- Context Injection Patterns (3 min read)
Module 3: Skills and Hooks as Instruction Engineering
Slash command skills and hooks are instruction documents, not convenience shortcuts. Write them with the same precision as your CLAUDE.md — because they extend it.
- Skills as Precision Prompts (4 min read)
- Hook-Based Instruction Injection (4 min read)
- Testing Agentic Instructions (4 min read)