The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external RST and Markdown documentation as its primary input, creating a surface for indirect prompt injection. Instructions embedded within documents being translated could potentially attempt to override agent logic or safety filters.
Ingestion points: Technical documentation files in the source repository ingested for translation (SKILL.md, Workflow Step 4).
Boundary markers: The instructions do not define explicit delimiters to separate untrusted document content from the agent's internal instructions, although they do provide strict rules for literal translation.
Capability inventory: The skill has the ability to write files to the project directory, execute Sphinx documentation builds, and configure GitHub Pages deployment when authorized (references/sphinx-publishing.md).
Sanitization: The workflow relies on mandatory manual review (Step 5) and fidelity checks to ensure instructions in the source text are translated as data rather than followed as commands.
[COMMAND_EXECUTION]: The skill utilizes a local utility script scripts/term_scan.py to identify technical terms and verify bilingual括注 (parenthetical annotations).
The script is a read-only tool using standard Python libraries (argparse, pathlib, re) to perform lexical matching against local glossary files.
It does not perform network requests or process untrusted code, and it provides specific exit codes for integration into pipelines.
[SAFE]: All external documentation links and repository references in the glossaries point to the vendor's own infrastructure or established, well-known project documentation sites (e.g., docs.kernel.org, docs.yoctoproject.org), which are considered safe references for technical context.

docs-translate-zh