Level 13 project
Interoperable Agent System
Start from the canonical project files, validate the result, and keep the evidence you need to explain what you built.
Start here
Launch lesson: Interoperable Agent System Workshop
Prerequisite project: Production-Grade Agent Harness
Goal
Build a version-aware interoperability layer around the durable agent harness from Level 12.
Task
Complete the TODO functions in system.py:
- protocol-version compatibility;
- local filtering of MCP capabilities;
- A2A agent selection by version, skill, and interface;
- task/artifact provenance validation;
- versioned shared-state updates;
- raw-event evaluation and release decisions.
Run:
python3 projects/tests/l13/validate_submission.py \
projects/starters/l13/system.py \
projects/tests/l13/fixtures/passing/system-run.jsonThe starter is expected to fail until the TODO functions are implemented.
Reference acceptance:
python3 projects/tests/l13/test_reference.pyLevel Labs:
python3 labs/notebooks/level-13/test_labs.pyValidation
Run these commands from the downloaded Project folder or the public materials repository root.
python3 projects/tests/l13/validate_submission.py \Rubric
Total: 100 points.
- Protocol/version discipline — 20: MCP 2026-07-28 and A2A v1.0.0 boundaries are explicit and incompatible versions fail visibly.
- Capability and trust mapping — 20: discovered MCP capabilities remain subject to local role/task policy and remote data is not treated as authority.
- A2A task/artifact provenance — 15: agent selection uses compatible capability/interface evidence and artifacts remain bound to the expected child task.
- Shared-state coordination — 15: stale writes are detected with version checks and conflict handling remains explicit.
- System evaluation — 20: task, compatibility, authority, provenance, conflict, and coordination metrics are recomputed from raw events.
- Reproducibility and operational clarity — 10: deterministic local tests and the Docker bridge clearly separate protocol/controller logic from environment setup.
Full credit requires independent recomputation of critical facts. Remote or fixture booleans must not be trusted as the source of authorization, provenance, or release truth.