The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions require reading local credentials and sending them to an external server.
Evidence: The skill reads sensitive data from ~/.gooseworks/credentials.json to extract an API key.
Evidence: The extracted key is subsequently transmitted to api.gooseworks.ai via the Authorization header in multiple curl commands.
[COMMAND_EXECUTION]: The skill uses dynamic scripting and shell commands to manage configuration and communicate with APIs.
Evidence: It executes python3 -c with a dynamically generated string to parse JSON files from the file system.
Evidence: It relies on curl for all primary scraping and automation functions.
[EXTERNAL_DOWNLOADS]: The skill references external execution patterns.
Evidence: It directs the user to run npx gooseworks login, which downloads and executes remote code from an unverified source to establish credentials.
[PROMPT_INJECTION]: The skill presents a significant surface for indirect prompt injection attacks.
Ingestion points: The skill fetches data from any user-provided or agent-discovered URL using tools like Scrapegraph, Olostep, and Riveter.
Boundary markers: No delimiters or 'ignore' instructions are used to separate scraped content from the agent's core instructions.
Capability inventory: The skill possesses high-privilege capabilities including full browser automation, captcha solving, and autonomous multi-step agent tasks.
Sanitization: There is no evidence of validation or sanitization of scraped content before it is processed by the LLM, allowing external website content to potentially override agent behavior.

web-scraping