Prompt Engineering
Advanced

Agentic Workflows

Design multi-step assistants that can inspect, decide, act, and verify.

45 min
3 sections
agents
planning
tools
1
2
3

01. Agents need operating rules

Section 1 of 3

An agent is an LLM loop with tools, memory, and objectives. The prompt should define scope, planning behavior, tool permissions, stop conditions, escalation conditions, and verification requirements.

text
Agent instructions:
- Work only inside the assigned repository.
- Inspect before editing.
- Prefer minimal changes.
- Run targeted tests after edits.
- Stop and ask for approval before deleting files, changing public APIs, or touching secrets.
- Report changed files, tests run, and remaining risk.
Back to Course