Skip to content

Library

The library is where course artifacts become reusable harness material. The repo already contains most of these artifacts inside project solution/ directories. This section makes them easier to find.

The artifacts come in two kinds. Composable parts (P01 through P07) stack into the harness.py you build across the course: a routing policy, a security profile, a critic, and so on. Decision rules (P08 through P10) are judgment you carry forward, like when to escalate models or when to index history instead of scanning it, even when they never become a line of harness.py.

Copy-Ready Categories

ArtifactUse it forSource project
Trace checklistDiagnosing a run from eventsP01
Routing policyChoosing a model before a task startsP02
MCP decision ruleDeciding when search tools earn their slotP03
Decomposition planSplitting large work into smaller runsP04
AGENTS.mdDurable repo memoryP05
Security profileRisk classification and confirmation policyP06
Critic rubricEvaluation and iterative refinementP07
Workflow skillBounded model-authored orchestrationP08
Escalation policySwitching models on evidenceP09
History index policyWhen to index agent history vs scan itP10

Start With These

The next useful extension is to extract the best artifacts into a dedicated library/ folder in the repo so learners can copy them without digging through project solutions.

Built as a friendly front door for the runnable OpenHands harness lab.