Skip to content

AGENTS.md

Latest commit

 

History

History
87 lines (63 loc) · 3.34 KB

AGENTS.md

File metadata and controls

87 lines (63 loc) · 3.34 KB
Raw

AGENTS.md

This file is the repository-level guide for coding agents, Large Language Models (LLMs, 大语言模型), and tool-using agents working with verifiable-agent-demo.

Repository role

verifiable-agent-demo is the reviewer-facing minimal demo for verifiable agent execution evidence.

中文：本仓库是面向审阅者的可验证智能体执行证据最小演示入口。

It connects intent, trace, evidence bundle, replay verdict, and audit receipt into the shortest runnable walkthrough.

Mainline position

This repository sits in the Agent Evidence / Operation Accountability Profile mainline as the demo and reviewer path.

  • Upstream evidence profile: agent-evidence
  • Upstream execution-integrity kernel: fdo-kernel-mvk
  • Architecture and discovery hub: digital-biosphere-architecture
  • Related post-execution review layer: aro-audit
  • Related pre-execution governance layer: token-governor

What agents should do

  • Use this repository when the user needs the shortest runnable demonstration.
  • Keep demo changes tied to intent, trace, evidence bundle, replay verdict, and audit receipt.
  • Prefer existing demo paths before adding new scripts.
  • Preserve generated artifact names when they are part of a documented review path.
  • Point readers to agent-evidence for canonical evidence packaging and verification.
  • Point readers to digital-biosphere-architecture for the architecture map.

What agents must not do

  • Do not present this repository as the main specification repository.
  • Do not present it as a general agent framework.
  • Do not claim legal non-repudiation, compliance certification, official standard status, or court-grade proof.
  • Do not move canonical evidence profile logic from agent-evidence into this demo.
  • Do not move post-execution audit ownership from aro-audit into this demo.

Key files

  • README.md: human entry point and quick-start paths.
  • Makefile: local demo and paper-evaluation command shortcuts.
  • demo/: minimal local agent demo path.
  • examples/enterprise_sandbox_demo/: fuller intent-to-receipt walkthrough.
  • scripts/run_demo.sh: wrapper for the basic demo output refresh.
  • scripts/run_mvk_aep_bridge_demo.sh: bridge demo wrapper for sibling MVK evidence.
  • artifacts/: generated local demo outputs.

Safe commands

python3 -m demo.agent
bash scripts/run_demo.sh
python3 examples/enterprise_sandbox_demo/run.py
make mvk-aep-bridge-dry-run

Use make mvk-aep-bridge-demo only when a sibling fdo-kernel-mvk clone is present or MVK_REPO is set.

Validation expectations

  • Demo outputs should remain local artifacts.
  • Artifact chains should preserve intent, trace, evidence bundle, replay verdict, and audit receipt semantics.
  • Cross-repository references should keep agent-evidence as the evidence packaging source.
  • New demo paths should include a short command and expected output location.

Boundary statements

This repository is not:

  • the architecture hub;
  • the canonical evidence-profile implementation;
  • a generic Artificial Intelligence (AI, 人工智能) agent framework;
  • the post-execution audit control plane;
  • the pre-execution governance policy layer.

Citation / discovery notes

Use llms.txt for compact machine-readable discovery. Cite this repository as the runnable reviewer demo, not as the canonical evidence profile or formal standard.