MANDATE (Multi-Agent Nominal Decomposition for Autonomous Task Execution) is a tolerance-based task specification framework for autonomous agent systems.
It produces governance-ready specification artifacts that separate:
- What constitutes acceptable success (the Anchor: minimum / target / constraints), from
- How to achieve it (multiple Courses of Action, COAs),
- With Search→Select→Trace provenance and optional Gap Analysis output when intent cannot be fully specified.
This repository is an implementation scaffold for the MANDATE paper (v1.0). It focuses on:
- JSON Schemas for mandate artifacts, 2) hashing + trace integrity utilities, and 3) a small CLI for validation.
schemas/— JSON Schemas for:mandate-as-code.schema.jsontrace-entry.schema.jsongap-report.schema.json
src/mandate/— Python utilities and a CLI:mandate validate examples/quarterly_report_mandate.json
examples/— example artifacts (mandate + gap report) and trace entriesdocs/— short specs and implementation notes
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
mandate validate examples/quarterly_report_mandate.json# Validate artifact schema and hashes
mandate validate <path/to/mandate.json>
mandate validate <path/to/gap.json>
# Compute hashes
mandate hash-anchor <path/to/mandate.json>
mandate hash-trace <path/to/trace_entry.json>
# Constraint grammar validation
mandate check-constraint "status == 'active' AND priority > 5"
mandate validate-constraints <path/to/mandate.json>- This is not an execution runtime. It does not run tools, enforce policy, or monitor runtime behavior.
- Constraint predicates follow a formal grammar (see
docs/artifact-spec.md). Theconstraints.pymodule provides parsing and validation. Translation to enforcement engines (OPA/Rego, Cedar) is a downstream responsibility.
versionfields are semantic and should be bumped when schemas change.- Hashes are SHA-256 over a deterministic JSON representation (see
docs/hashing.md).
Apache 2.0. See LICENSE for details.
