Skip to content
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
720 changes: 67 additions & 653 deletions README.md

Large diffs are not rendered by default.

152 changes: 101 additions & 51 deletions README.zh-CN.md
Original file line number Diff line number Diff line change
Expand Up @@ -16,85 +16,131 @@ Agent Evidence 是 Digital Biosphere Architecture 的具体 execution-evidence

它把 agent/runtime 执行打包成可离线验证的证据 bundle。它不是完整的架构总仓,不是 audit control plane,也不只是 tracing 或 logging。要看系统上下文,请先去 [digital-biosphere-architecture](https://github.com/joy7758/digital-biosphere-architecture);要走最短演练路径,请看 [verifiable-agent-demo](https://github.com/joy7758/verifiable-agent-demo);要做执行后审阅,请看 [aro-audit](https://github.com/joy7758/aro-audit)。

## 角色
## 研究定位

`agent-evidence` 是具体的 execution-evidence 入口,用来把 agent/runtime 的执行过程打包成可移植、可离线验证的证据包。
Agent Evidence 只做一个收敛主张:执行证据与操作问责,是可治理 AI
系统的一等验证边界。

## 不是这个仓库
本仓库是 Digital Biosphere Architecture 中的具体 execution-evidence
入口。它把一次 operation 打包为可移植、可独立复核、可离线验证的对象化证据;它是方法入口,不是架构总仓,不是 audit control plane,也不是泛化的
agent governance 平台。

- 不是完整的架构总仓
- 不是 audit control plane
- 不只是 tracing 或 logging
- 不是 walkthrough demo
- 不是 execution-integrity kernel
## 当前规范包

## 从这里开始

- architecture context -> [digital-biosphere-architecture](https://github.com/joy7758/digital-biosphere-architecture)
- 当前主 package -> `spec/execution-evidence-operation-accountability-profile-v0.1.md`、`schema/execution-evidence-operation-accountability-profile-v0.1.schema.json`
- 当前可运行入口 -> [examples/README.md](examples/README.md)、[demo/README.md](demo/README.md)、`agent-evidence validate-profile <file>`
- 历史脉络 -> [docs/lineage.md](docs/lineage.md)
- walkthrough -> [verifiable-agent-demo](https://github.com/joy7758/verifiable-agent-demo)
- post-execution review -> [aro-audit](https://github.com/joy7758/aro-audit)

## 当前 v0.1 package

当前主 package surface 是
当前 canonical package 是
`Execution Evidence and Operation Accountability Profile v0.1`。

它冻结在 GitHub Release `v0.2.0` 中。

当前 package DOI:https://doi.org/10.5281/zenodo.19334062

该 release 内部冻结的 package 版本仍是 `v0.1`。

当前 v0.1 路径从这里开始:
冻结信息:
- GitHub Release:`v0.2.0`
- DOI:[10.5281/zenodo.19334062](https://doi.org/10.5281/zenodo.19334062)
- 该 release 内部冻结的 package 版本仍是 `v0.1`

核心入口:
- 规范:`spec/execution-evidence-operation-accountability-profile-v0.1.md`
- Schema:`schema/execution-evidence-operation-accountability-profile-v0.1.schema.json`
- Validator CLI:`agent-evidence validate-profile <file>`
- Examples:[examples/README.md](examples/README.md)
- Demo:[demo/README.md](demo/README.md)
- 样例:`examples/README.md`
- Demo:`demo/README.md`
- reviewer-facing 高风险入口:`docs/high-risk-scenario-entry.md`
- 状态与验收:`docs/STATUS.md`、`docs/ACCEPTANCE-CHECKLIST.md`
- 提交交付:`submission/package-manifest.md`、`submission/final-handoff.md`
- 投稿交付:`submission/package-manifest.md`、`submission/final-handoff.md`

实现说明:JSONL、SQLite 和 PostgreSQL 后端仍然可用,但它们从属于本仓库作为 evidence-entry 的定位。
## FDO 对外文稿入口

![存储](https://img.shields.io/badge/storage-JSONL%20%7C%20SQLite%20%7C%20Postgres-0a7b83)
如果当前目标是 FDO Testbed 注册、外联或项目插入申请,先看:

### Minimal v0.1 walkthrough
- `docs/fdo-mapping/fdo-operation-evidence-profile-registration-pack.md`
- `submission/fdo-testbed-registration-draft.md`
- `submission/peter-sven-outreach-draft.md`
- `submission/ldt4ssc-ds4sscc-module-pitch.md`

```bash
python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
```
当前对外命名关系:

验证最小 valid / invalid 样例:
- `FDO_OPERATION_EVIDENCE_PROFILE_V0_1` = operation-level evidence profile
- `ARO_AUDIT_PROFILE_V1` = audit-facing sibling profile

```bash
agent-evidence validate-profile examples/minimal-valid-evidence.json
agent-evidence validate-profile examples/invalid-missing-required.json
agent-evidence validate-profile examples/invalid-unclosed-reference.json
agent-evidence validate-profile examples/invalid-policy-link-broken.json
```
## Minimal v0.1 walkthrough

运行最小 demo:
python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

```bash
python3 demo/run_operation_accountability_demo.py
```
agent-evidence validate-profile examples/minimal-valid-evidence.json
agent-evidence validate-profile examples/invalid-missing-required.json
agent-evidence validate-profile examples/invalid-unclosed-reference.json
agent-evidence validate-profile examples/invalid-policy-link-broken.json

预期结果:
agent-evidence validate-profile examples/valid-high-risk-payment-review-evidence.json
agent-evidence validate-profile examples/invalid-high-risk-unclosed-reference.json
agent-evidence validate-profile examples/invalid-high-risk-policy-link-broken.json

python3 demo/run_operation_accountability_demo.py

预期结果:
- valid 样例返回 JSON,其中 `"ok": true`
- 每个 invalid 样例返回 JSON,其中 `"ok": false`,并给出一个主错误码
- demo 会把工件写到 `demo/artifacts/`,最后输出一行 `PASS` 摘要

已知环境说明:

- 仓库 `.venv` 在 Python 3.14 下跑更大范围测试时,可能出现一条 `langchain_core` warning;它不影响最小 profile、validator 或 demo 路径

## 论文台账

下面这些文件就是当前项目与论文状态的统一台账入口:

- 项目状态与里程碑台账:`docs/STATUS.md`
- 旗舰论文工作日志:`paper/flagship/WORKLOG.md`
- 稿件基线说明:`submission/manuscript-baselines.md`
- claims-to-evidence 映射表:`paper/flagship/13_claims_to_evidence_map.md`
- validation 结果总表:`paper/flagship/18_validation_results_table.md`

## 稿件口径纪律

不要混写不同稿件表面。

- `B1-minimal-frozen`:`Execution Evidence and Operation Accountability Profile v0.1`;主张 = 最小验证边界
- `B4-high-risk-current-main`:reviewer-facing 的高风险入口;更适合未来 high-risk / compliance-interface 稿件
- `B2-extended-middle`:除非整篇重写,否则继续停放
- `B3-aep-live-chain`:保留为历史 AEP runtime-evidence 表面

## 本仓库已经建立了什么

本仓库当前已经建立了:
- execution evidence 与 operation accountability 的最小 profile
- 带显式 error code 的 profile-aware validator
- 单链路 demo
- reviewer-facing 的场景切片
- 能把论文主张压到具体仓库资产上的证据表面

## 范围纪律

这个仓库是:
- 具体 execution-evidence 入口
- 最小验证边界规范包
- validator / specimen / demo 的落地表面

这个仓库不是:
- 架构总仓
- audit control plane
- walkthrough demo
- execution-integrity kernel
- 泛化 agent governance 平台
- manifesto 仓库

## 架构导航

- 架构总入口 -> `digital-biosphere-architecture`
- 最短演练路径 -> `verifiable-agent-demo`
- 执行后审阅入口 -> `aro-audit`
- EDC Java spike 入口 -> `docs/edc/edc_extension_minimal_structure.md`
- 历史脉络图 -> `docs/lineage.md`

## 下一步重点

- external-context evidence
- third-party checker
- introduction / discussion / conclusion 的连续 manuscript assembly

## 历史脉络

历史上的 `Execution Evidence Object`、较早的 `Agent Evidence Profile` 命名、旧版 FDO mapping 表述以及会议样品说明仍然保留在仓库里,但它们已经不是当前主入口。历史脉络请统一看 [docs/lineage.md](docs/lineage.md)。
Expand Down Expand Up @@ -129,6 +175,10 @@ Tracing 和 logs 主要帮助操作者检查一次运行。Agent Evidence 把运
这个仓库当前实现了 bundle、manifest、signatures 和 offline verification
这些步骤。外部 anchoring 不在 AEP v0.1 的范围内,默认也不会启用。

可选的 trust binding 和 manifest 签名不是一回事。当前 profile 中它表现为
`validation.trust_bindings[]`,只用于指向外部验证来源,不替代本地签名。
当前 validator 只检查本地 target 和 digest 是否一致,不校验外部系统本身。

该工具包现在支持两种存储模式:

- 仅附加本地 JSONL 文件
Expand Down
10 changes: 10 additions & 0 deletions agent_evidence/integrations/__init__.py
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,8 @@
)
from .langchain import (
EvidenceCallbackHandler,
LangChainAdapter,
LangChainArtifacts,
evidence_from_langchain_event,
record_langchain_event,
)
Expand All @@ -16,12 +18,20 @@
exported_trace_summary,
install_openai_agents_processor,
)
from .openai_compatible import (
OpenAICompatibleAdapter,
OpenAICompatibleArtifacts,
)

__all__ = [
"AgentEvidenceTracingProcessor",
"build_erc8004_validation_stub",
"build_fdo_stub",
"EvidenceCallbackHandler",
"LangChainAdapter",
"LangChainArtifacts",
"OpenAICompatibleAdapter",
"OpenAICompatibleArtifacts",
"evidence_from_langchain_event",
"evidence_from_openai_agents_span",
"evidence_from_openai_agents_trace",
Expand Down
Loading