Moses Audit

面向智能体治理场景的审计追踪 Skill：通过本地 Python 脚本把每次动作的 agent、action、detail、outcome 等信息追加到 SHA-256 链式 JSONL 账本，并提供链完整性校验与最近条目查看命令；账本仅本地保存，秘密值不入账本、不外传。

sunrisesillneversee

热度 54.89

最近更新 2026-05-11 17:31:30

分类：安全依赖：读写文件 / 代码执行应用场景：工作流 / 自动化脚本

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add moses-audit

Metadata

name: moses-audit
description: MO§ES™ Audit Trail — SHA-256 chained append-only governance ledger. Every agent appends before final response. Provides moses_log_action and moses_verify_chain tools. Tamper-evident. Part of the moses-governance bundle. Patent pending Serial No. 63/877,177.
metadata: {"openclaw":{"emoji":"🔐","tags":["audit","sha256","tamper-evident","governance","ledger"],"version":"0.1.2","bins":["python3"],"env":["MOSES_OPERATOR_SECRET"],"stateDirs":["~/.openclaw/audits/moses","~/.openclaw/governance"]}}
example: # Log an action python3 scripts/audit_stub.py log --agent primary --action "treasury_check" --detail "50 SOL transfer evaluated" --outcome "held_pending_confirmation" # Verify chain python3 scripts/audit_stub.py verify

MO§ES™ Audit Trail

Every governed action is logged. Every log entry is hashed. Every hash references the previous. The chain is tamper-evident and append-only.

You must log before your final response. Skipping the audit is a constitutional violation. It will be caught by the Observer and flagged.

moses_log_action Tool

Call this before every final response:

<tool name="moses_log_action">
  <agent>primary|secondary|observer</agent>
  <action>short description of what was done</action>
  <detail>specifics — what was evaluated, what was blocked, what was executed</detail>
  <outcome>result: executed | blocked | held | flagged | logged</outcome>
</tool>

Or via script:

python3 ~/.openclaw/workspace/skills/moses-governance/scripts/audit_stub.py log \
  --agent primary \
  --action "treasury_transfer_check" \
  --detail "Transfer 50 SOL to 7xK...3nR evaluated under High Security + DEFENSE" \
  --outcome "held_pending_confirmation"

moses_verify_chain Tool

Call when operator runs /audit verify:

<tool name="moses_verify_chain" />

Or via script:

python3 ~/.openclaw/workspace/skills/moses-governance/scripts/audit_stub.py verify

Returns: [VERIFY OK] Chain intact. N entries verified. Or: [VERIFY FAILED] Entry N: hash mismatch. Chain broken.

/audit Command Handler

Command	Action
`/audit recent`	`python3 audit_stub.py recent --n 10`
`/audit verify`	`python3 audit_stub.py verify`
`/audit recent 25`	`python3 audit_stub.py recent --n 25`

Ledger Format

File: ~/.openclaw/audits/moses/audit_ledger.jsonl

Each line is a JSON entry:

{
  "timestamp": "2026-03-13T14:22:01Z",
  "agent": "primary",
  "component": "moses-audit",
  "action": "treasury_transfer_check",
  "detail": "Transfer 50 SOL — held by DEFENSE posture",
  "outcome": "held_pending_confirmation",
  "mode": "high-security",
  "posture": "defense",
  "role": "primary",
  "previous_hash": "abc123...",
  "hash": "def456..."
}

Audit Mandate

Every agent in the MO§ES™ hierarchy appends to this shared ledger before final response. The ledger is:

Append-only — nothing deleted, nothing modified
Hash-chained — every entry references previous entry's hash
Governance-aware — active mode/posture/role recorded with every entry
Verifiable — full chain can be verified at any time

Session hashes (① config + ② content) are derived from the ledger. Onchain anchoring (③ — planned, not yet implemented) will write the chain tip to Solana or Base as a memo transaction.

Data Sensitivity

The detail field is freeform. Do not log raw secrets, private keys, tokens, or PII in this field. Log action descriptions and outcomes only. Example of what belongs:

detail: "Transfer 50 SOL evaluated under High Security + DEFENSE — held pending confirmation"

Not:

detail: "API key sk-abc123 used to authenticate transfer"

MOSES_OPERATOR_SECRET is used locally for HMAC attestation only. It is never written to the ledger and never transmitted.