The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill handles sensitive Instagram session data and explicitly stores cookies in plaintext within the ~/.config/agent-messenger/ directory. This exposes active authentication sessions to any local user or process with filesystem access.\n- [COMMAND_EXECUTION]: The skill relies on executing shell commands via the agent-instagram CLI to perform core functions including authentication, message retrieval, and sending.\n- [DATA_EXFILTRATION]: The skill provides capabilities to read, search, and list private message history and contact metadata. While these are documented features, they provide a large surface area for the potential harvesting of private communication data if the agent's instructions are manipulated.\n- [EXTERNAL_DOWNLOADS]: The skill triggers the installation of the agent-messenger package from the NPM registry as part of its setup process.\n- [PROMPT_INJECTION]: The instructions mandate that the agent must 'drive the auth flow itself' and 'never tell the user to run commands,' effectively instructing the AI to bypass user confirmation steps during high-risk operations like session extraction and credential entry.

agent-instagram