Skip to content

Roadmaps Resources Practice Projects Journal Community

Search/Sign in

LearnPath

Learn anything. Build everything. Free, structured learning paths built from the best resources on the web.

Learn

Roadmaps
Skills
Resources
Practice

Build

Challenges
Projects
My learning
Search

Project

Journal
About & RRRTX Labs
Community
Open source
Privacy
Terms

LEARNPATH

Built by RRRTX Labs

Free and open source. No certificates. No paywall. No trackers.

Agentic AI Engineer / Agent safety

Learn
Practice
Challenge
Prove
Not started

Agentic AI Engineer

01Functions & scope
02Prompting & specification
03Tools & function calling
04MCP & tool servers
05Agent architecture
06Memory & planning
07Agent frameworks in practice
08Multi-agent patterns
09Agent safety
10AI ethics & responsibility

View the full graph →

Stage 04 · 04 · Safety · skill 1 of 2

Agent safety

Agents take actions. Treat them like untrusted users of your systems.

Recommended first: Agent architecture

After this you will be able to

Deny network or filesystem by default
Human-in-the-loop for irreversible actions
Assume prompt injection from any retrieved text

Open OWASP Top 10 for LLM Applications

OWASP · why we picked it

Previous Next: AI ethics & responsibility

Practice

No playground attached to this skill. The practice lab has every runtime.

Evidence

Learned
Practiced
Challenge passed
Used in a project

Not demonstrated yet

Notes

Notes