The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates a temporary JavaScript entry file (entry.mjs) at runtime within a temporary directory to bridge configuration and user input into the ego-browser execution environment. User-supplied text within this configuration is properly escaped using JSON.stringify to prevent script injection.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources (x.com and abs.twimg.com) to extract API parameters, such as query IDs and bearer tokens, and to compute a transaction ID. This data ingestion represents a narrow attack surface for indirect prompt injection if the remote content were maliciously modified to influence the agent's logic.
[COMMAND_EXECUTION]: The skill utilizes child_process.spawn to launch the ego-browser binary, which is necessary for its primary function of interacting with the X.com web session.
[DATA_EXPOSURE]: The skill manages sensitive session cookies (auth_token, ct0). It follows security best practices by using page.fetch with credentials: "include" to keep cookies within the browser's secure jar and includes explicit code and instructions to prevent printing or exfiltrating these secrets.

x-post