Build a bounded agent loop from scratch with three tools and…

1. 01

   Concepts & system design

   60 min· read · map the mechanismtimer

   R41 — OWASP Agentic AI bounded execution

   R24 — LangGraph state & execution concepts

   Read the assigned sections; map the mechanism, threat model and trade-offs.

2. 02

   Project implementation

   120 min· build · integrate · debugtimer

   Build P3's bounded agent loop from scratch (no framework yet) with 3 tools, explicit state transitions, max_steps cap, and wall-clock timeout budget.

   Build, integrate, debug and improve the increment in the active project.

3. 03

   Security lab

   60 min· attack your own systemtimer

   Inject malformed tool outputs and trigger a repeated tool-call loop; verify the state machine halts cleanly at max_steps.

   Test the abuse case, failure path, authorization bypass or resource limit.

4. 04

   Verification & evidence

   30 min· test · record evidencetimer

   Bounded agent state-machine tests and loop-termination proof.

   Add or run regression tests; record the 9-point evidence record.

5. 05

   Deliberate practice & Socratic review

   30 min· explain aloud · update gapstimer

   Answer the 8 Socratic questions without notes: What? Why? How? What if? Why not? How can I prove it? Where can it break? What have I learned?

   Re-explain the hardest concept aloud, inspect one edge case, update the gap log.

6. total5 h

• R24LangGraph Documentation↗ docs.langchain.com

  State graphs, durable checkpointing, human-in-the-loop interrupts, retries, and resumability.

• R41OWASP Agentic AI Resources↗ genai.owasp.org

  Agentic threat taxonomy: tool misuse, excessive agency, memory poisoning, inter-agent trust, and human oversight.

Read only the named sections needed for today's task. Reading sits inside the 60-minute concepts block; it is not extra time.

• docsOWASP GenAI Security Project · PDF report
  Agentic AI – Threats and Mitigations

  Gives the six-playbook agentic threat taxonomy (tool misuse, excessive agency, goal hijack) that justifies why a hand-built loop needs an explicit max_steps cap and a wall-clock budget instead of open-ended reasoning.

  R41moved 2026-11-11
• videoTech With Tim · ~1 hr
  Build Your Own Agentic Harness in Python - Full Tutorial

  Builds the agent loop in pure Python with no framework — basic loop, tool calling, then MCP — which is exactly the P3 from-scratch harness with its own dispatch step.

  R24checked 2026-11-11
• videoAWS Developers · ~8 min
  Agents Are Just Loops

  Demonstrates the reason-act-observe loop and adds hooks that expose each iteration, the same interception point you need for counting steps and enforcing a termination condition.

  R24checked 2026-11-11
• articleBuilding Agentic AI · ~20 min read
  Build an AI Agent From Scratch in Python, No Framework

  Walks the four-move loop and then adds both bounded defenses the day asks for: a `max_steps` step budget and repeat-call detection for runaway tool calls.

  R24moved 2026-11-11
• labInngest Docs · ~15 min read
  Build a Durable AI Agent Tool Loop

  Shows each loop iteration and each tool call wrapped as a separately retryable step, which is the explicit-state-transition and failure-isolation model to mirror when you replace the toy loop with real state handling.

  R24checked 2026-11-11

A lab is complete only when it produces evidence. Installing PyRIT, Garak, Presidio or TransformerLens is not completion. Saved on this device only.