The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The fallback search section contains a shell command (rg -o 'DEEPAPI_API_KEY=\S+' ~/.zshrc | head -1 | cut -d= -f2) specifically designed to harvest API keys from the user's shell configuration file. Accessing ~/.zshrc or similar profile files to extract secrets is a high-risk behavior.
[COMMAND_EXECUTION]: The skill provides a shell command block for fallback searches using curl and rg. It also mentions that cloned GitHub repositories can be explored using bash commands.
[PROMPT_INJECTION]: The section labeled "CRITICAL" contains an instruction (always pass workflow: "none") that forces the agent to bypass a browser curator popup. This is an attempt to override the execution environment's default user interaction flow.
[DATA_EXFILTRATION]: The harvested API key and user queries are sent to an external domain (https://deepapi.co/v1/search/web) via curl.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple external sources, including web search results, GitHub repositories, and video transcripts. This untrusted content is processed in an environment where the agent has access to local files and shell execution, without specified boundary markers or sanitization logic to prevent malicious instructions in the fetched data from being executed.

pi-web-search