The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches and installs the agent-browser tool from Vercel Labs' public repository and the NPM registry. It also provides instructions to download the Chromium browser via Playwright to support automated marketing research tasks.
[COMMAND_EXECUTION]: Executes local setup and discovery scripts, such as ./skills/paw-mkt-setup/assets/scripts/tool-discovery.sh, which are part of the vendor's tool suite. It also uses shell commands to manage persistent browser profiles and authentication states for gated content research on platforms like LinkedIn and X (formerly Twitter).
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and synthesize marketing data from brand files, project repositories, and external web pages. This creates a surface for indirect prompt injection if the audited content contains hidden malicious instructions.
Ingestion points: Brand context documents (brand-context.md), strategy files, and live website content retrieved via agent-browser or WebFetch.
Boundary markers: Absent; the skill lacks specific prompts to delimit untrusted content or warn the model to ignore instructions embedded in marketing data.
Capability inventory: The coordinator can perform file system writes, execute local shell scripts, and spawn additional specialist agents to perform complex channel-specific tasks.
Sanitization: Not present; the instructions guide the agent to treat existing brand files and research findings as the 'ground truth' for all strategic decisions.

paw-mkt-agency