The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses curl to fetch web content through r.jina.ai, a well-known service for Markdown conversion. This is a standard and legitimate practice for translation agents processing external URLs.
[COMMAND_EXECUTION]: The workflow involves executing a local Python script (scripts/fix_punctuation.py) provided with the skill. The script performs regex-based punctuation correction on intermediate files using standard libraries.
[PROMPT_INJECTION]: As the skill processes untrusted user data from URLs and text inputs for translation, it has an inherent indirect prompt injection surface.
Ingestion points: Content is retrieved from user-provided URLs via curl or read from input files.
Boundary markers: The skill uses a multi-stage workflow (Analysis, Draft, Critique) to process data, though explicit delimiters like XML tags are not strictly enforced in the final translation prompts.
Capability inventory: The agent can execute shell commands (curl) and run local Python scripts.
Sanitization: The skill does not perform explicit sanitization of the scraped Markdown content before passing it to translation subagents.

translate-polisher