The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides the agent with capabilities to fetch and process data from arbitrary external URLs and search results. This content is untrusted and could contain malicious instructions designed to manipulate the agent's behavior or override its safety guidelines.
Ingestion points: scripts/webtools.py via subcommands scrape, interactive, search, and crawl which ingest external web data.
Boundary markers: The skill lacks explicit boundary markers or delimiters to wrap the scraped content, increasing the risk that the agent might interpret data as instructions.
Capability inventory: The skill allows network access to api.zoodata.ai and execution of the webtools.py CLI script.
Sanitization: The CLI script outputs raw data from the API (JSON or Markdown) without performing sanitization to strip potential injection payloads.
[CREDENTIALS_UNSAFE]: The skill requires a ZOODATA_API_KEY for authentication. The instructions guide the agent to store this key in a local file at ~/.zoodata/config.json. Although the instructions include best practices for securing the file (using chmod 700 and umask 077), the storage of a sensitive API key in a known file path creates a risk of credential exposure if the agent's environment is accessed by a malicious actor.

web-extract