The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes instructions in references/shared-patterns.md to install the agent-browser tool from GitHub (github.com/vercel-labs/agent-browser) and the NPM registry. As Vercel Labs is a trusted organization, these remote references are considered safe and intended for the skill's operation.
[COMMAND_EXECUTION]: In references/shared-patterns.md, the skill provides instructions to execute local scripts for tool discovery and profile management (e.g., tool-discovery.sh, chrome-profiles.sh). These scripts are expected components of the broader marketing suite and are used for environment configuration.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its automated scraping of social media content from TikTok, Instagram, and YouTube for influencer vetting. This allows untrusted third-party content to enter the agent's context.
Ingestion points: Content scraped from public social media profiles and search result pages via agent-browser get text body in references/influencer-identification.md.
Boundary markers: No boundary markers or instructions to ignore embedded commands within the scraped text were identified.
Capability inventory: The skill can perform file system writes (saving campaign deliverables), network operations (via agent-browser), and local command execution (as described in shared-patterns.md).
Sanitization: There are no explicit instructions for the agent to sanitize, validate, or escape the content ingested from external websites before processing it.

paw-mkt-influencer