Agent governance suite: policy-based tool gating (allow/deny/ask with wildcards and priorities), a structured JSONL audit trail, and per-agent token quotas against the host token meter, with state under $DSH_HOME/gov.
Install
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:863683348/dsh-gov
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
This plugin publishes its README in Chinese only.
权限、审计、成本 —— DSH 进企业前必须有的东西,做成一个插件。策略门禁(allow/deny/ask)、结构化审计日志、按 agent 的 token 配额,状态持久化在 $DSH_HOME/gov/。
为什么需要它
企业引入 DSH 时第一波问题永远是:
- 谁能让 agent 跑高危工具? 策略门禁:按工具/agent/工作区规则,allow / deny / ask,通配符 + 优先级,默认放行、命中即裁决、平局 fail-closed(deny > ask > allow)。
- 出事了怎么追溯? 每次工具调用(决定 + 结果)都写入结构化 JSONL 审计日志(
audit.jsonl),可查询、可导出、可汇总。 - 钱花哪去了? 接宿主
tokenMeter,按 agent 累计 token 用量,周期配额(day/week/month/total),超限注入上下文警告,可一键重置。
安装
dsh plugin --profile <profile> add dsh-gov
用法(模型侧)
工具 gov,action 一览:
| action | 说明 |
|---|---|
status |
治理总览:规则数、审计事件汇总(allow/deny/ask/error/ok)、配额用量 |
policy_list / policy_add / policy_remove |
规则管理(tool 支持 * 通配;priority 高者胜;平局 fail-closed) |
audit_query / audit_export |
查询审计(limit/agent/tool/since 过滤);导出日志路径 |
quota_get / quota_set / quota_reset |
按 agent id(或 global)的 token 预算 |
示例:
gov policy_add tool="pwsh*" policyAction=ask reason="shell commands require approval" priority=10
gov quota_set id=alice quotaLimit=100000 period=day
gov audit_query decision=deny limit=20
配置
| key | 默认 | 说明 |
|---|---|---|
root |
$DSH_HOME/gov |
治理数据目录(gov.json + audit.jsonl) |
defaultPeriod |
day |
新配额默认周期 |
defaultLimit |
0 |
默认配额(0 = 不限;>0 时启用 step 计量) |
prestepWarn |
true |
超限时向模型上下文注入警告 |
sectionOrder |
5 |
提示词段落顺序 |
设计
- 纯逻辑分层:
lib/policy.js(规则引擎)、lib/audit.js(审计模型)、lib/quota.js(配额计算)零依赖可单测;lib/index.js负责接线(tools/pre-execute门禁、tools/result观察、agent/pre-step计量)。 - 全局而非会话:状态和日志在 DSH home,跨工作区一致。
- fail-safe:计量失败、持久化失败都不阻塞 agent;策略引擎平局默认 fail-closed。
- 审计日志是纯追加 JSONL,可被任何合规管道消费(二期:导出 SOC2 素材、多租户隔离)。
测试
node test/policy.test.mjs && node test/audit.test.mjs && node test/quota.test.mjs
FAQ
- 默认放行还是默认拒绝? 默认
allow(与 DSH 一致);需要收紧就加规则,平局时 fail-closed(deny > ask > allow)。 - 审计日志会记录工具参数吗? 不会。只记工具名、agent、工作区、决定、理由、结果(成功/失败)——不含参数内容。
- 配额按什么计? 宿主
tokenMeter的totalTokens(当前请求+响应压力),按 agent id 累计到周期。 - 计量失败会卡住 agent 吗? 不会,fail-safe:计量/持久化异常时跳过该步。
License
MIT
Links
More in this category
toby-bridges/api-relay-audit★ 819
Runs local security audits of AI API relays and LLM proxies from DeepSeek Harness, producing Markdown reports for prompt injection, model substitution signals, tool-call rewriting, error leakage, stream integrity, and profile-gated Web3 risks.
howmp/dsh-pentest★ 330
Authorized pentest mode for DeepSeek Harness — exploration chain, assets and findings with a Web view.
SeaOf0/dsh-redteam-model★ 138
Authorized-security DSH collection: nine work modes (redteam coordinator, pentest, code audit, binary analysis, attack-defense, AV evasion, incident response, cloud security, CTF solving) and fifteen runtime plugins, managed from a settings page with one-click deploy, install, update and uninstall.
PerryLink/dsh-auto-review★ 124
Second-model auto-review on the approval answerer chain: a read-only reviewer subagent returns structured allow/deny verdicts with reasons, fail-closed by default.
PerryLink/dsh-permission-rules★ 81
Claude Code-style declarative permission rules: ordered allow/deny/ask YAML rules matching tool names, arguments, workspace paths, and agent identity on the tools/pre-execute waterfall, with full session-log audit, dry-run mode, and hot reload.
PensiveFei/dsh-secure-audit★ 72
Read-only security and compliance plugin for DeepSeek Harness: prompt-injection detection, Chinese-PII redaction, and a local configuration audit with redacted, reproducible reports.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.