The Agent Skills Directory

[COMMAND_EXECUTION]: The file references/shared-patterns.md contains instructions for the agent to install external dependencies using shell commands such as npx skills add and npm install -g. While these target trusted sources, they represent arbitrary command execution for environment setup.
[COMMAND_EXECUTION]: In references/influencer-identification.md, the skill defines shell commands for the agent-browser tool that incorporate dynamic, user-influenced parameters (e.g., {niche-keyword}). This represents a potential surface for command injection if the underlying platform does not properly sanitize these inputs.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download and installation of the agent-browser tool from Vercel Labs' official GitHub repository.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from external websites (TikTok, Instagram, YouTube) via the agent-browser tool.
Ingestion points: Web content is retrieved using agent-browser get text body in references/influencer-identification.md.
Boundary markers: None are explicitly used to delimit external content from system instructions.
Capability inventory: The agent has the ability to write files (saving research and reports) and perform network operations via the browser tool.
Sanitization: No specific sanitization or filtering logic is defined for the ingested web data.

bmad-agent-marketing-influencer