The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external HTML files during the translation workflow, creating an attack surface for instructions embedded in document content.
Ingestion points: The sub-agent defined in references/translate-document.md reads source HTML files from the .aigne/doc-smith/dist/ directory.
Boundary markers: The agent uses a structured checklist and specific role-play instructions ('专业的多语言本地化翻译专家') to guide the sub-agent, though the run_in_background: true flag limits immediate user oversight of task output.
Capability inventory: The sub-agent has access to Write, Edit, and Bash tools, which could be abused if an injection attack succeeds.
Sanitization: Instructions explicitly command the agent to not translate script tags, HTML attributes, or code blocks, and to verify the resulting HTML structure before saving.
[COMMAND_EXECUTION]: The skill executes multiple shell commands and a local Node.js script for file management and building documentation artifacts.
Evidence: Uses find, wc, cat, mkdir, rm -rf, and cp to manage status files and synchronize assets within the workspace.
Evidence: Executes node skills/doc-smith-build/scripts/build.mjs to rebuild navigation files, which is a local script execution consistent with the vendor's documentation ecosystem.
Evidence: Uses the shasum command to calculate document hashes for incremental build logic.

doc-smith-localize