Skip to content

cacohe/agent-language-policy

v0.1.0Apache-2.0

Keep human decision content in a chosen language while preserving technical literals.

Agent Language Policy

English | 简体中文 | 繁體中文 | 한국어

Keep the explanations that developers must review in a chosen language while preserving code, commands, paths, APIs, identifiers, logs, and original errors.

Support

  • Languages: Simplified Chinese (zh-CN), Traditional Chinese (zh-TW), and Korean (ko-KR). The default is zh-CN.
  • Agents: Codex, Claude Code, Cursor, Gemini CLI, and GitHub Copilot.
  • Delivery: a portable Agent Skill, persistent project or user instructions, a CLI, and a Markdown validator.
  • Runtime: Node.js 20 or later, with no runtime dependencies.

The npm package is published as agent-language-policy. The portable Skill is available in the GitHub v0.1.0 release. The Cursor Marketplace submission is under review, Gemini CLI Gallery automatic indexing is pending, and OpenAI and Claude directory submissions have not been made.

Install the portable Skill

The community skills CLI can install human-decision-language into a supported Agent:

npx skills add cacohe/agent-language-policy --skill human-decision-language

This command installs only the portable Skill. It does not install persistent instructions, create configuration, or provide the Markdown validator. Install the policy CLI separately from npm or source when you need those capabilities.

Install the policy CLI from npm

Run the published CLI without a permanent global installation:

npx --yes --package agent-language-policy@latest -- agent-language-policy doctor
npx --yes --package agent-language-policy@latest -- agent-language-policy install --agent codex --scope project --language zh-CN --dry-run
npx --yes --package agent-language-policy@latest -- agent-language-policy install --agent codex --scope project --language zh-CN

Run these commands from the project that should receive the persistent rule. Replace codex with another supported Agent and zh-CN with zh-TW or ko-KR as needed.

Policy boundaries

  • Human decision content—such as designs, plans, reviews, progress updates, risks, and approval requests—uses the configured language.
  • Technical literals retain English or their original language.
  • Project-owned README files, product documentation, UI copy, code comments, commit messages, pull requests, changelogs, and public API documentation follow repository and audience conventions.
  • Natural-language instructions cannot guarantee perfect model compliance. The validator provides an additional check for Markdown artifacts written to the repository.

Install from source

Clone this repository, verify it, then run the CLI directly:

git clone https://github.com/cacohe/agent-language-policy.git /path/to/agent-language-policy
cd /path/to/agent-language-policy
npm test
cd /path/to/your-project
node /path/to/agent-language-policy/bin/agent-language-policy.mjs doctor

Preview and install a persistent project rule:

node /path/to/agent-language-policy/bin/agent-language-policy.mjs install --agent codex --scope project --language zh-CN --dry-run
node /path/to/agent-language-policy/bin/agent-language-policy.mjs install --agent codex --scope project --language zh-CN

Replace codex with claude-code, cursor, gemini-cli, or github-copilot. Replace zh-CN with zh-TW or ko-KR as needed. Cursor user-level rules require the manual step printed by the CLI.

Repository layout

PathResponsibility
skills/Portable Agent Skills and self-contained references
policy/Canonical language data used by the CLI
src/cli/Command parsing and orchestration
src/configuration/Configuration, profile loading, and policy rendering
src/installation/Agent adapters and managed-block lifecycle
src/validation/Markdown prose extraction and language validation
bin/Executable entry point
scripts/Static and package-release verification
test/Unit, integration, structure, and package tests
.github/workflows/Cross-platform continuous integration

Operate and remove

node /path/to/agent-language-policy/bin/agent-language-policy.mjs status --agent codex --scope project
node /path/to/agent-language-policy/bin/agent-language-policy.mjs validate docs/plans/feature.md --language zh-CN
node /path/to/agent-language-policy/bin/agent-language-policy.mjs uninstall --agent codex --scope project --dry-run
node /path/to/agent-language-policy/bin/agent-language-policy.mjs uninstall --agent codex --scope project

Uninstalling removes only the managed block.

Trust and compatibility

Review the Skill and generated instruction block before installation. The installer creates a backup before changing an existing file and never grants tools or shell permissions. The file mappings for all five Agents are covered by automated tests, and the npm package and GitHub release are public. Cursor review and Gemini CLI Gallery indexing remain external pending steps; OpenAI and Claude directory submissions have not been made.

See Architecture, SECURITY.md, and CHANGELOG.md for design, project policies, and release history. Report ordinary problems through GitHub Issues; report vulnerabilities through the private process in SECURITY.md.

Development

npm test
npm run verify