The Agent Skills Directory

[COMMAND_EXECUTION]: The research-playbook.md file defines several agent-browser commands to automate web research, including taking screenshots and extracting text from competitor sites and industry benchmark resources. It also provides installation commands using npm and npx.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to download and install the agent-browser tool from a trusted source repository (github.com/vercel-labs/agent-browser) if it is not already available in the execution environment. Documentation describes this neutrally as a setup requirement.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data retrieved from external websites via browser automation, creating a vulnerability surface.
Ingestion points: Content extracted from third-party websites (competitors, blogs, benchmark sites) via agent-browser and user-provided local brand/product context files.
Boundary markers: The skill does not define specific delimiters or explicit instructions to the agent to ignore potential commands embedded in the fetched research data.
Capability inventory: The skill has the ability to write deliverables to the local file system (via the saving protocol) and execute further browser-based research commands.
Sanitization: No evidence of automated sanitization or validation of external web content is present in the skill's reference files before the data is processed or included in recommendations.

paw-mkt-retention