The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing the agent-browser skill from the vercel-labs GitHub repository if it is not already present. This is used for automated content research. The repository belongs to a well-known organization and is used for its intended purpose of enhancing the skill's research capabilities.
[COMMAND_EXECUTION]: The skill provides instructions to execute shell commands for research and installation purposes. This includes using agent-browser to fetch data from various websites like Google, Reddit, and Quora, as well as using npm or npx to install necessary tools and dependencies.
[DATA_EXPOSURE]: This skill follows a research-heavy workflow that involves processing data from the user's project files, including strategic brand context and SOSTAC plans. While this is necessary for its function as a marketing specialist, it involves reading sensitive business strategy files.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from external websites via agent-browser (e.g., from Reddit, Quora, or competitor blogs).
Ingestion points: Browser-based research commands in references/content-research.md (Google PAA, Reddit, Quora, competitor blogs).
Boundary markers: Not explicitly defined for the data returned from browser sessions.
Capability inventory: Network access (via agent-browser), file system access (reading brand context, writing marketing content).
Sanitization: No specific sanitization or validation of the fetched HTML/text content is mentioned before it is synthesized into content strategies.

paw-mkt-content