The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The configuration file scripts/config.py contains a hardcoded API key (IMGUR_CLIENT_ID) for the Imgur service, which is used for anonymous uploads.\n- [DATA_EXFILTRATION]: The skill automatically transmits local images and videos to an external, non-whitelisted third-party domain, Imgur (https://api.imgur.com/3/image), as part of the content publication pipeline in scripts/publish.py and scripts/api_client.py. This behavior results in local user data being stored on a third-party service to facilitate the Instagram Graph API's requirement for public URLs.\n- [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection by fetching untrusted external content and providing dangerous tools to the agent: 1. Ingestion points: scripts/comments.py (lists comments), scripts/messages.py (reads DMs), and scripts/hashtags.py (reads external posts). 2. Boundary markers: The instructions do not define delimiters or provide warnings to the agent regarding instructions embedded in the fetched Instagram content. 3. Capability inventory: The skill includes tools for public interaction, including scripts/publish.py (content creation), scripts/comments.py (replying to and deleting comments), and scripts/messages.py (sending direct messages). 4. Sanitization: There is no sanitization or filtering of the fetched content before it is processed by the agent.\n- [COMMAND_EXECUTION]: The skill relies on and exposes numerous shell commands for managing account setup, authentication, and service orchestration, which increases the risk of command injection or misuse if the agent is manipulated.

instagram