Labs Companion

The labs companion

The L.A.B. Simulator: Learn → Apply → Break. Auto-graded, decision-driven labs, every one offline and without an AWS account.

Open the labs on GitHub ↗ · the pathway · grade one in the Arena

Each lab has three phases, and the Break phase is the point:

	Phase	What happens
L	Learn	A mental model, a diagram, and one decision you must make — with the trade-offs laid out and the answer withheld. Your decision changes what you build.
A	Apply	Implement against a spec. Public checks you can see, hidden checks on submit. Every check says what it teaches when it fails.
B	Break	Your working solution meets the cases that end real runs: BaseException from a library, a chunk bigger than the whole budget, a summariser that returns more than it was given. Survive them.

Start in 60 seconds

In Codespaces, open one and type lab next. Locally:

python labs/runner/labctl.py list          # the catalog
python labs/runner/labctl.py next          # what you can start right now
python labs/runner/labctl.py start PDL-01  # copies a starter into your workspace
python labs/runner/labctl.py show  PDL-01  # the brief
python labs/runner/labctl.py run   PDL-01  # public checks
python labs/runner/labctl.py submit PDL-01 # public + hidden, records the attempt
python labs/runner/labctl.py break PDL-01  # now survive the Break phase
python labs/runner/labctl.py progress      # per-track progress bars

The catalog

10 labs built · 4h 40m of hands-on work · every one runs offline, with no AWS account.

📋 Product & PDLC

_{Decide what to build before building it}

Lab	Difficulty	Time	Teaches	Break phase
PDL-01 · Agent, workflow, or script	easy	25m	autonomy ladder, control flow, scoping	✓

🔁 Agent Loop

_{The loop, by hand, until nothing is magic}

Lab	Difficulty	Time	Teaches	Break phase
AGL-01 · Dispatch a tool call	easy	20m	tool dispatch, toolUse / toolResult, unknown-tool handling	✓
AGL-02 · Close the loop	easy	25m	message history, role alternation, parallel tool calls	✓
AGL-03 · Stop the loop	medium	30m	termination, step budget, oscillation	✓

🔧 Tools

_{Schemas, contracts and honest failure}

Lab	Difficulty	Time	Teaches	Break phase
TOOL-03 · Fail honestly: no bare empties	medium	25m	failure honesty, empty vs absent, tool contracts	✓

🧠 Memory

_{What the agent keeps, and what it drops}

Lab	Difficulty	Time	Teaches	Break phase
MEM-01 · A buffer that cannot overflow	medium	30m	context budget, eviction policy, summarisation threshold	✓

📚 Retrieval

_{Grounding you can actually verify}

Lab	Difficulty	Time	Teaches	Break phase
RET-05 · Citations that survive verification	medium	35m	citation integrity, grounding, provenance	✓

🕸️ Multi-Agent

_{Topologies, and what they cost}

Lab	Difficulty	Time	Teaches	Break phase
MAS-02 · Cost a topology before you build it	medium	30m	handoff multiplier, context re-transmission, merge cost	✓

🔬 Evaluation

_{Proving it works, and blocking it when it does not}

Lab	Difficulty	Time	Teaches	Break phase
EVAL-03 · A gate that can say no	medium	30m	release gating, absolute vs average metrics, threshold governance	✓

🚀 Production

_{Deployed, observable, reversible}

Lab	Difficulty	Time	Teaches	Break phase
PROD-02 · Failover that cannot be silent	medium	30m	model failover, silent degradation, observable fallback	✓

More labs are specified and open for contribution: the full pathway is the 41-lab map, and how to author one is the guide.

Which module teaches each lab

Lab	Difficulty	Time	Learn it in
AGL-01 · Dispatch a tool call	easy	20 min	Module 05 · The Agent Loop: No Framework to Strands
AGL-02 · Close the loop	easy	25 min	Module 05 · The Agent Loop: No Framework to Strands
AGL-03 · Stop the loop	medium	30 min	Module 05 · The Agent Loop: No Framework to Strands
EVAL-03 · A gate that can say no	medium	30 min	Module 13 · Agentic QA and Evaluation
MEM-01 · A buffer that cannot overflow	medium	30 min	Module 09 · LLM Memory Mechanics
MAS-02 · Cost a topology before you build it	medium	30 min	Module 07 · Multi-Agent Patterns with Strands
PDL-01 · Agent, workflow, or script	easy	25 min	Module 00 · Agentic Foundations
PROD-02 · Failover that cannot be silent	medium	30 min	Module 14 · End-to-End Production Pipeline
RET-05 · Citations that survive verification	medium	35 min	Module 10 · RAG, OpenSearch and LiteLLM
TOOL-03 · Fail honestly: no bare empties	medium	25 min	Module 06 · Strands Foundations: Tools, Memory and MCP

Drills, when you have ten minutes

Post /drill AGL-101 in the Simulator Arena and a bot grades it in the thread; progress lands on the Scoreboard.

Three ways to run a lab

	Setup	You get
🚀 Codespaces	One click	Everything — hidden checks, progress, the DAG, briefs beside your editor
💻 Local	git clone, Python 3.11+	The same, on your machine
💬 Simulator Arena	None. A comment box	Public + Break checks, graded by a bot in the thread

---

The course companion · Teach it: the cohort kit · Exercises with answers

_{✏️ This page is generated from labs/README.md and the lab manifests, an edit made here is replaced at the next sync.}