The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted markdown files sequentially, which creates a vulnerability surface for indirect prompt injection if source files contain adversarial instructions.
Ingestion points: Processes documentation content from docs/docs/en and docs/docs/includes/en inside SKILL.md.
Boundary markers: Absent. There are no boundary delimiters or specific directives instructing the agent to ignore instructions embedded within the text files being translated.
Capability inventory: Executes automated shell commands via git checkout and performs multi-file write modifications to the workspace and the mkdocs.yml file.
Sanitization: Absent. Content is processed and written back to the workspace without parsing or sanitization filters to strip potential prompt injections.
[COMMAND_EXECUTION]: Instructs the agent to execute a git command (git checkout -b docs-translation-<language>) to create a new branch for the translations.

mkdocs-translations