The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted web content from search engines (Google, Bing, Yandex) and discovery tools, which creates a surface for indirect prompt injection attacks where malicious instructions hidden in search results could attempt to influence agent behavior.\n
Ingestion points: Web search results fetched via bdata search and bdata discover (described in SKILL.md and references/patterns.md).\n
Boundary markers: Absent. The instructions do not define clear delimiters or warnings for the agent to disregard instructions within the scraped content.\n
Capability inventory: The skill utilizes the bdata CLI for network access, performs file system writes, and uses shell processing utilities.\n
Sanitization: The skill employs jq to extract structured data fields and grep to filter out block-page content, providing a degree of technical sanitization.\n- [COMMAND_EXECUTION]: The skill provides complex shell command patterns for batch processing and result filtering.\n
Evidence: references/patterns.md contains shell loops and pipes using xargs, bash -c, and md5sum to automate search tasks.\n- [DYNAMIC_EXECUTION]: The skill documents patterns for dynamically constructing shell commands based on variable inputs.\n
Evidence: Shell scripts in references/patterns.md use string interpolation to build command arguments and file paths. The documentation demonstrates proper quoting (e.g., "$q", "$url") to mitigate injection risks.

search