Guide 03AdvancedPlaceholderSecurity and containment
Contain Agent Actions: Permissions, Prompt Injection, and Human Approval
Enforce boundaries in the environment, not the prompt, and test them with adversarial content.
- Written guide
- Labs
Dev Guides
Each guide pairs a written lesson with hands-on labs that fail on purpose, an anchored assessment and the material to teach it to a room. Companion videos and short clips accompany the guides that have them.
Sample figures from placeholder guides

Featured · Guide 01 · Intermediate
PlaceholderImplement a bounded tool loop, diagnose a seeded failure, and prove the final state instead of trusting a success message.
Browse the library
Filter by skill, level and format. Guides list only what they include.
1 of 3 guides
Clear 2 filters
Guide 03AdvancedPlaceholderSecurity and containment
Enforce boundaries in the environment, not the prompt, and test them with adversarial content.
How every guide is built
Every lab is executed before publishing, with exact versions, commands and outputs recorded.
Each guide includes a seeded failure, its symptom, the diagnosis and how to verify the fix.
Objectives map to a practical task and an anchored rubric, not a recall quiz.
An instructor kit covers run of show, 12 and 100 learner formats, dry runs and clearance.
Published guides meet all four. The placeholder guides show sample sections only.