The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the internet, creating a surface for indirect prompt injection attacks.
Ingestion points: The skill fetches content from arbitrary external URLs using the fetch command in harness/scripts/web-crawl.py.
Boundary markers: There are no explicit instructions or delimiters defined to separate the fetched content from the agent's primary instructions, nor are there warnings to the agent to ignore embedded instructions in the scraped text.
Capability inventory: The skill writes processed markdown to the local raw/ directory and is designed to feed data into other agent workflows such as /propose or ingest.
Sanitization: While the skill uses regex to convert HTML to markdown, it does not perform semantic sanitization to strip or neutralize natural language instructions contained within the fetched content.
[COMMAND_EXECUTION]: The skill relies on the execution of a local Python script (harness/scripts/web-crawl.py) to perform its primary functions, including network requests and file writing.
[EXTERNAL_DOWNLOADS]: The skill is explicitly designed to perform network operations to download content from external web servers. While this is its intended purpose, it involves connecting to potentially untrusted third-party domains.

web-crawl