The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow executes a Python script (scripts/content_ingester.py) to process user-provided URLs or file paths, and utilizes shell commands like mkdir for output directory management.
[EXTERNAL_DOWNLOADS]: The skill utilizes the requests library and the DrissionPage browser automation tool to fetch content from arbitrary external URLs provided by the user.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core ingestion workflow.
Ingestion points: Untrusted data enters the agent context through scripts/content_ingester.py and is temporarily stored in config/raw_content.txt.
Boundary markers: SKILL.md includes explicit instructions stating that URL content, webpage bodies, and documents are 'untrusted input' that must not 'override system rules' or be 'treated as commands'.
Capability inventory: The skill possesses capabilities for subprocess execution (via the ingestion script), directory creation (mkdir), and file writing (Write tool).
Sanitization: The ingestion script uses html2text to strip HTML noise and BeautifulSoup for basic metadata extraction, though it does not perform semantic validation of the content.

knowledge-absorber