The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/chrome-profile-discovery.py programmatically searches for and identifies paths to sensitive browser data, specifically targeting Cookies and Login Data files for Google Chrome and Microsoft Edge across Windows, macOS, and Linux. This script reports the existence and location of these files to the agent context.
[CREDENTIALS_UNSAFE]: The documentation in references/browser-tools.md instructs the agent to save authenticated session states, which contain plaintext session tokens, to the local project directory (e.g., .auth/session.json). Storing session tokens in plaintext poses a high risk of credential theft if the filesystem is compromised or if the agent is manipulated into exfiltrating these files.
[EXTERNAL_DOWNLOADS]: The skill requires the installation and use of an external third-party CLI tool, agent-browser, from the NPM registry. This tool is granted deep access to the user's browser profiles to perform authenticated web research, which creates a significant security dependency on a non-vendor-controlled tool.
[INDIRECT_PROMPT_INJECTION]: The skill's primary research workflows involve ingesting content from untrusted external sources, including social media feeds (Instagram, TikTok, LinkedIn) and the general web via the agent-browser and Exa tools. There are no explicit instructions or mechanisms mentioned for sanitizing or delimiting this ingested content before it is processed and synthesized into strategic reports, creating a surface for indirect prompt injection attacks.

paw-cra-agent-strategist