The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install external software globally if not already present. Specifically, it executes npm install -g agent-browser && npx playwright install chromium to enable its automated research capabilities. While these are standard tools for browser automation, installing global packages from the internet is a noteworthy behavior.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection. It is designed to perform deep research by scraping content from untrusted external sources including Reddit, Quora, G2 reviews, and Google search results. This untrusted data is then interpolated into the agent's context to generate strategic recommendations and marketing plans. While the skill includes human-in-the-loop review gates between phases, the ingestion of arbitrary web content into the reasoning chain remains a risk.
Ingestion points: Web scraping via agent-browser from sources like reddit.com, quora.com, g2.com, and google.com (specified in references/auto-discovery.md).
Boundary markers: The skill uses structured markdown phase documents but does not explicitly detail sanitization or specific delimiters for the external content within the prompts.
Capability inventory: The skill has capabilities to read/write files in the brand workspace, execute shell commands for research tools, and perform extensive network operations.
Sanitization: No specific sanitization or filtering of external content is mentioned beyond the synthesis step.
[COMMAND_EXECUTION]: The skill utilizes the agent-browser CLI to perform research tasks, involving complex command sequences to open URLs, wait for network activity, and extract text or screenshots.

paw-mkt-sostac