ADVANCED · 8 MODULES · 64 LESSONS
Agent Security and Sandboxing
Make sure an agent with real tools can't be tricked into deleting, spending, or leaking data, even when the model or the person typing to it is hostile.
- For you if
- You're building an agent that takes real actions and worry that hostile input could turn it against you.
- By the end
- You'll know how to stop prompt injection, give each tool the least access it needs, isolate its execution, and test your defenses with attacks.
- You need
- Python 3.14 and a basic grasp of tool-calling agents. No API key; Docker for one example only.

Curriculum
01Por qué la seguridad de agentes
- Module 1: Why Agent Security
- From talking to acting: why an agent with tools is dangerous
- An agent's threat model
- Excessive agency
- The OWASP LLM Top 10 (and the agentic Top 10) as a framework
- How this differs from classic app security
- Anatomy of an incident: how an agent attack chains together
- Mini-project: a vulnerable Reservo agent
02Inyección de prompt
- Module 2 Introduction: Prompt Injection
- What prompt injection is
- Direct injection: when the attacker is the user
- Indirect injection: the poisoned data
- Why the LLM doesn't separate instructions from data
- Prompt defenses: delimiters and instruction hierarchy
- Why no prompt defense is enough: you have to limit agency
- Mini-project: inject and defend
03El contrato de herramientas y el mínimo privilegio
- Module 3 Introduction: the tool contract and least privilege
- A tool's contract and its JSON Schema
- Least privilege: only the tools you need
- Narrow tools vs. a too-powerful tool
- Separating reads from writes and marking the destructive
- The tool allowlist and safe dispatch
- Designing a safe tool: bounding `run_query`
- Mini-project: a least-privilege registry
04Sandboxing de la ejecución de herramientas
- Module 4 Introduction: sandboxing tool execution
- Why confine a tool's execution
- Subprocess: isolating from the agent's process
- Limits with `resource`: CPU and memory
- The timeout that kills a hung process
- No network and an isolated temp dir
- What a sandbox protects and what it does not
- Mini-project: a sandboxed code tool
05Modelos de permiso y human-in-the-loop
- Module 5 Introduction: permission models and human-in-the-loop
- Not all actions are equal: reversible vs destructive
- The three levels: auto / ask / deny
- Human-in-the-loop: the approval gate
- Capability-based permissions, not ambient access
- The audit log of every tool call
- Designing an agent's permission policy
- Mini-project: a permission-gated agent
06Guardrails de entrada y salida
07Secretos, aislamiento y radio de explosión
- Module 7 Introduction: secrets, isolation, and blast radius
- Secrets outside the prompt: the model never sees them
- Minimal-scope credentials per tool
- Network isolation and host allowlists
- The blast radius and how to shrink it
- Defense in depth: the layers together
- Monitoring and anomaly detection on the audit log
- Mini-project: shrink the blast radius
08Módulo 8 (proyecto): endurece un agente con herramientas
Reviews
What students say
These reviews are from enrolled students who completed at least 50% of the course. We moderate reviews only on content grounds (spam, offensive language, personal data), never for being critical or negative.
No approved reviews yet.
Be the first to share your experience!