aump-py

Python SDK and validator for the Agentic User Mandate Protocol.

This package provides:

• bundled AUMP v0.1 JSON Schemas;
• mandate schema and semantic validation;
• evidence event schema validation;
• evidence event semantic validation against mandate reference, retention, and required event policy;
• deterministic allowed, denied, and requires_escalation action evaluation;
• MCP, A2A, and UCP/AP2 bridge validation helpers;
• a small in-process runtime for examples and agent applications;
• a CLI for schema validation and action evaluation.

Install In This Workspace

uv sync

Global Install Smoke

After the package is published, the CLI should work with either uv tool or standard Python installers:

uv tool install aump
aump --help

CLI

Validate a mandate:

uv run aump validate mandate ../conformance/fixtures/mandates/marketplace-buyer.valid.json

Validate an evidence event:

uv run aump validate evidence-event ../conformance/fixtures/events/deal-accepted.valid.json

Validate evidence semantics against a mandate:

uv run aump validate-evidence \
  --mandate ../conformance/fixtures/mandates/marketplace-buyer.valid.json \
  --event ../conformance/fixtures/events/deal-accepted.valid.json

Evaluate an action:

uv run aump evaluate-action \
  --mandate ../conformance/fixtures/mandates/marketplace-buyer.valid.json \
  --action ../conformance/fixtures/actions/accept-ping-pong.allowed.json

Python

from aump import evaluate_action
from aump.policy import parse_datetime

result = evaluate_action(mandate, action, now=parse_datetime("2026-04-25T18:00:00Z"))
assert result["decision"] in {"allowed", "denied", "requires_escalation"}

Test

The tests prefer the sibling conformance/fixtures corpus when this workspace is cloned end to end. CI falls back to the pinned tests/fixtures/conformance snapshot so this repo can validate itself.

uv run ruff check .
uv run pytest