The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/web_search.py prints the full request headers to standard output within the ve_request_with_apikey function. This includes the Authorization header containing the plaintext WEB_SEARCH_API_KEY, leading to sensitive credential exposure in the agent's execution logs.
[CREDENTIALS_UNSAFE]: The instructions in SKILL.md direct the agent to ask users for sensitive credentials (WEB_SEARCH_API_KEY, VOLCENGINE_ACCESS_KEY, VOLCENGINE_SECRET_KEY) and write them to an environment file within the local workspace, which may be accessible to other processes or subsequent agent tasks.
[COMMAND_EXECUTION]: The skill operates by executing a local Python script (scripts/web_search.py) through a shell command, passing user-provided queries as arguments.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests unsanitized summaries from external websites and presents them to the agent for response generation.
Ingestion points: Search result summaries from the Result.WebResults API field are extracted in scripts/web_search.py.
Boundary markers: No delimiters or 'ignore' instructions are used when passing the web content to the agent.
Capability inventory: The agent has the capability to execute local scripts and write to the filesystem.
Sanitization: No filtering or sanitization is performed on the external web content beyond basic whitespace stripping.

web-search