The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from external X post URLs.
Ingestion points: pull.py fetches HTML content from the publish.x.com/oembed endpoint.
Boundary markers: Extracted data is structured in TSV format (swipe.tsv) and eventually rendered into Markdown swipe files (swipe.md) using specific headers and bullet points to delimit entries.
Capability inventory: The skill performs statistical calculations and text extraction using regex. It does not use eval(), exec(), or subprocesses to execute the ingested text as code.
Sanitization: pull.py uses html.unescape and multiple re.sub patterns to strip HTML tags and isolate plain text before storage and analysis.
[SAFE]: Network activity is restricted to the official publish.x.com/oembed endpoint, which is a well-known service for embedding public posts and does not require authentication or user session data.
[SAFE]: The skill's instructions and implementation explicitly forbid requesting user credentials or performing unauthorized scraping, adhering to platform safety guidelines.
[SAFE]: File system operations are restricted to specific local directories (e.g., ~/.claude/x/) and use standard data formats (TSV/Markdown) for persistence.

x-viral