The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the agent-browser dependency from Vercel Labs' official GitHub repository (https://github.com/vercel-labs/agent-browser). This is a well-known and trusted source for agent tooling.
[COMMAND_EXECUTION]: The skill executes shell commands via agent-browser and npx to perform market research, competitive analysis, and journalist outreach discovery. These commands are scoped to the intended functionality of the skill.
[DATA_EXPOSURE]: The instructions direct the agent to read local strategic files (e.g., ./brands/{brand-slug}/brand-context.md, ./brands/{brand-slug}/sostac/03-strategy.md) to ground its PR recommendations in the specific business context. This represents normal operating behavior for a marketing specialist skill.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the web (Google News, LinkedIn, etc.) via agent-browser.
Ingestion points: agent-browser get text body is used to fetch live news and search results.
Boundary markers: The instructions guide the agent to "Extract" specific information from the fetched body, though explicit delimiters for the external content are not mandated in the prompt snippets.
Capability inventory: The agent can write resulting strategy and PR files to the local directory ./brands/{brand-slug}/campaigns/pr/.
Sanitization: There is no explicit sanitization step mentioned for the fetched web content, but the risk is assessed as low given the standard research use case.

marketing-pr