The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves fetching and processing content from arbitrary external URLs and PDFs, which represents a significant surface for indirect prompt injection attacks.
Ingestion points: The skill ingests data via scripts/fetch.sh, scripts/fetch_local.py, scripts/fetch_weixin.py, and scripts/fetch_feishu.py.
Boundary markers: SKILL.md contains explicit instructions to treat fetched content as untrusted data, disregard embedded priority overrides, and warn the user if prompt-like instructions are detected.
Capability inventory: The skill can execute shell commands, perform network requests via curl, write files to disk, and use system utilities like pdftotext.
Sanitization: Content is cleaned and converted to Markdown (stripping HTML executable tags), but the skill relies on agent instructions to handle textual injection attempts.
[COMMAND_EXECUTION]: The skill orchestrates its operations by executing several local bash and Python scripts provided within the skill package, such as fetch.sh and fetch_local.py.
[DYNAMIC_EXECUTION]: The references/read-methods.md file contains a pattern for executing Python code directly from the command line (python3 -c) to handle PDF text extraction as a fallback method.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch web content and encourages the installation of various third-party packages (e.g., playwright, beautifulsoup4, requests, marker-pdf) from official registries. It also includes an optional feature to send URLs to third-party services like defuddle.md and r.jina.ai for extraction.

read