The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides instructions in references/shared-patterns.md for accessing sensitive browser profiles and session tokens. It encourages the use of persistent Chrome profiles (e.g., ~/.linkedin-profile, ~/.twitter-profile) and includes commands to save authentication states to local files like ./my-auth.json. The skill's own documentation explicitly warns that these files "contain sensitive session tokens in plaintext," which creates a risk of credential exposure if the environment is compromised.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of external software. It provides commands in references/shared-patterns.md and references/research-playbook.md to install agent-browser and Playwright dependencies using npm and npx. It fetches code from Vercel Labs' official GitHub repository.
[COMMAND_EXECUTION]: The skill utilizes shell commands for environment setup and research operations. It references local scripts for tool discovery and Chrome profile identification, and provides numerous command-line templates for browser automation using the agent-browser tool.
[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection. It is designed to fetch and process text content from arbitrary external URLs (competitor sites, blogs, documentation) via browser automation using agent-browser get text body.
Ingestion points: references/research-playbook.md fetches content from external URLs into the agent's context for research purposes.
Boundary markers: No delimiters or instructions to ignore embedded commands were found in the provided templates.
Capability inventory: The skill has access to shell execution via agent-browser and local file-writing protocols.
Sanitization: No evidence of input validation or sanitization of the scraped web content was identified.

paw-mkt-retention