The Agent Skills Directory

PROMPT_INJECTION (LOW): The skill is vulnerable to Indirect Prompt Injection (Category 8). It fetches content from arbitrary external websites and passes it directly into an LLM prompt without sanitization or boundary markers.\n
Ingestion points: Untrusted web content is ingested via curl in several sections, such as the 'Page Analysis' and 'Security Scan' blocks.\n
Boundary markers: Absent. The $CONTENT and $HEADERS variables are interpolated directly into prompts without XML tags, delimiters, or instructions to ignore embedded commands.\n
Capability inventory: The agent has access to curl, bash, and the gemini CLI tool to perform further network and file operations.\n
Sanitization: None. The skill performs no filtering or escaping of the fetched HTML before it is processed by the AI.\n- EXTERNAL_DOWNLOADS (SAFE): The skill performs downloads from user-specified URLs as part of its primary purpose (web crawling). While an automated scanner flagged curl | python3 as a remote code execution pattern, analysis of the source code confirms this is a false positive. The skill uses python3 -c to execute a static, embedded Python script that treats the downloaded content as data (stdin), not as executable code.\n- COMMAND_EXECUTION (LOW): The skill utilizes shell pipes and python3 -c for data processing and text extraction. These commands are static and do not incorporate untrusted external input into the command string itself.

spider