The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes a 'Live Website URL Audit' capability that ingests content from public URLs using browser automation or fetch tools. This introduces a surface for indirect prompt injection, where malicious instructions embedded in a scanned website could attempt to influence the agent's behavior during a session.
Ingestion points: The skill uses agent-browser and WebFetch to process external URLs as described in references/shared-patterns.md.
Capability inventory: The skill has the ability to write files, install packages (npm, npx), and perform browser automation.
Boundary markers: The skill does not explicitly define delimiters or warnings for the agent when processing this untrusted external data.
Sanitization: No specific sanitization or filtering logic for fetched web content is documented.
[EXTERNAL_DOWNLOADS]: The instructions guide the agent to install the agent-browser tool from Vercel Labs' official GitHub repository. This is a reference to a well-known and trusted service provider and is a standard part of the skill's setup for research tasks.
[COMMAND_EXECUTION]: The skill executes local setup and discovery scripts, such as tool-discovery.sh, to identify the local environment's capabilities. These are intended for diagnostic and configuration purposes within the marketing suite.
[CREDENTIALS_UNSAFE]: The skill manages persistent sessions for social media platforms by saving authentication states and browser profiles to local files (e.g., my-auth.json). While these files contain sensitive session tokens, the skill includes best-practice documentation advising the user to secure these files and exclude them from version control.

paw-mkt-social