AI 治理 DeepSeek Harness Adapter
This skill lets DeepSeek Harness (dsh) use the original claude-for-legal-ZH/ai-governance-legal content without requiring Claude Code slash commands.
Source Files
- Domain root:
ai-governance-legal
- Domain profile template and shared rules:
ai-governance-legal/CLAUDE.md
- Original skills directory:
ai-governance-legal/skills
- Original plugin description: AI 治理实务:对拟议的 AI 应用场景进行分类登记、依据适用监管体系开展算法安全评估与科技伦理审查、审查 AI 供应商条款中的训练数据使用和责任条款、保持 AI 使用政策与实践同步。
Path Resolution
The repository-relative paths above resolve against the claude-for-legal-ZH repository root:
- If the current workspace contains the repository, resolve them against the repository root directly.
- Otherwise (user-level install), run
cat ~/.dsh/legal-zh/repo once to get the absolute repository path recorded by scripts/install-dsh.sh, then prefix the relative paths with it.
How To Use
- Read
ai-governance-legal/CLAUDE.md before substantive work.
- Select the closest original skill from the list below, then read its
SKILL.md.
- Follow that skill's workflow, translating Claude Code slash-command wording into dsh actions and natural conversation.
- If multiple original skills apply, execute them in the order implied by the workflow and merge the result.
- Do not run Claude-specific plugin commands. Ignore Claude hooks. Use dsh filesystem, shell, web, and MCP tools for local files, verification, document rendering, and user-visible output.
Configuration Compatibility
The original project stores setup profiles under ~/.claude/plugins/config/.... For DeepSeek Harness, use this order:
- If a populated Claude profile exists, read it as the user's existing practice profile.
- Otherwise use or create
~/.dsh/legal-zh/ai-governance-legal/CLAUDE.md for dsh-specific setup.
- If the selected skill requires setup and the profile still contains
[PLACEHOLDER], run the domain's cold-start-interview workflow in conversation before producing customized legal work.
When an original instruction says to run /ai-governance-legal:some-command, interpret that as: load ai-governance-legal/skills/some-command/SKILL.md and perform the workflow in dsh.
Legal Retrieval MCP
If the active dsh profile mounts the chineselaw or yuandian MCP servers (see INSTALL_DSH.md), prefer mcp__chineselaw__* and mcp__yuandian__* tools for statute, case, and regulatory verification. Without them, mark time-sensitive legal facts as requiring verification before reliance.
Available Original Skills
ai-inventory, aia-generation, cold-start-interview, customize, matter-workspace, policy-monitor, policy-starter, reg-gap-analysis, use-case-triage, vendor-ai-review
Legal Output Rules
- Treat all output as lawyer-review draft work, not legal advice replacing professional judgment.
- Mark uncertain legal citations or case references as requiring verification unless verified from a reliable source in this session.
- For current law, regulatory updates, case retrieval, filing requirements, deadlines, or other time-sensitive legal facts, verify with current sources before relying on them.
- Preserve the original workflow's escalation, approval, confidentiality, and source-labeling requirements.
1---2name: chinese-legal-ai-governance3description: Use when the user needs Chinese legal work in the AI 治理 domain: AI 应用登记、算法安全评估、科技伦理审查、生成式 AI 合规、AI 供应商审查、AI 治理. This is a DeepSeek Harness (dsh) adapter for claude-for-legal-ZH/ai-governance-legal; it routes natural-language requests to the original domain CLAUDE.md and skills/*/SKILL.md workflows.4---56# AI 治理 DeepSeek Harness Adapter78This skill lets DeepSeek Harness (`dsh`) use the original `claude-for-legal-ZH/ai-governance-legal` content without requiring Claude Code slash commands.910## Source Files1112- Domain root: `ai-governance-legal`13- Domain profile template and shared rules: `ai-governance-legal/CLAUDE.md`14- Original skills directory: `ai-governance-legal/skills`15- Original plugin description: AI 治理实务:对拟议的 AI 应用场景进行分类登记、依据适用监管体系开展算法安全评估与科技伦理审查、审查 AI 供应商条款中的训练数据使用和责任条款、保持 AI 使用政策与实践同步。1617## Path Resolution1819The repository-relative paths above resolve against the `claude-for-legal-ZH` repository root:2021- If the current workspace contains the repository, resolve them against the repository root directly.22- Otherwise (user-level install), run `cat ~/.dsh/legal-zh/repo` once to get the absolute repository path recorded by `scripts/install-dsh.sh`, then prefix the relative paths with it.2324## How To Use25261. Read `ai-governance-legal/CLAUDE.md` before substantive work.272. Select the closest original skill from the list below, then read its `SKILL.md`.283. Follow that skill's workflow, translating Claude Code slash-command wording into dsh actions and natural conversation.294. If multiple original skills apply, execute them in the order implied by the workflow and merge the result.305. Do not run Claude-specific plugin commands. Ignore Claude hooks. Use dsh filesystem, shell, web, and MCP tools for local files, verification, document rendering, and user-visible output.3132## Configuration Compatibility3334The original project stores setup profiles under `~/.claude/plugins/config/...`. For DeepSeek Harness, use this order:35361. If a populated Claude profile exists, read it as the user's existing practice profile.372. Otherwise use or create `~/.dsh/legal-zh/ai-governance-legal/CLAUDE.md` for dsh-specific setup.383. If the selected skill requires setup and the profile still contains `[PLACEHOLDER]`, run the domain's `cold-start-interview` workflow in conversation before producing customized legal work.3940When an original instruction says to run `/ai-governance-legal:some-command`, interpret that as: load `ai-governance-legal/skills/some-command/SKILL.md` and perform the workflow in dsh.4142## Legal Retrieval MCP4344If the active dsh profile mounts the `chineselaw` or `yuandian` MCP servers (see `INSTALL_DSH.md`), prefer `mcp__chineselaw__*` and `mcp__yuandian__*` tools for statute, case, and regulatory verification. Without them, mark time-sensitive legal facts as requiring verification before reliance.4546## Available Original Skills4748`ai-inventory`, `aia-generation`, `cold-start-interview`, `customize`, `matter-workspace`, `policy-monitor`, `policy-starter`, `reg-gap-analysis`, `use-case-triage`, `vendor-ai-review`4950## Legal Output Rules5152- Treat all output as lawyer-review draft work, not legal advice replacing professional judgment.53- Mark uncertain legal citations or case references as requiring verification unless verified from a reliable source in this session.54- For current law, regulatory updates, case retrieval, filing requirements, deadlines, or other time-sensitive legal facts, verify with current sources before relying on them.55- Preserve the original workflow's escalation, approval, confidentiality, and source-labeling requirements.