The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill requires the user to provide a GitHub Personal Access Token (PAT) as a runtime variable. Although the instructions include explicit mitigations—such as a directive not to log, echo, or store the token—the manual handling of raw secrets by an AI agent creates a risk of accidental exposure in conversation logs, shell history, or process lists when constructing commands.
[COMMAND_EXECUTION]: The skill uses curl commands to transmit user data and the GitHub PAT to the vendor's API. This is the intended functionality, but executing shell commands with sensitive variables can lead to credential leakage if the agent is not strictly constrained.
[EXTERNAL_DOWNLOADS]: The installation process fetches the skill configuration from the official vendor domain at immersivecommons.com. This is a standard procedure for this vendor's infrastructure.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via the ic_leaderboard_get_board functionality.
Ingestion points: The agent fetches and processes a JSON response containing a list of members, including handle and name fields which are user-controlled on GitHub.
Boundary markers: The instructions do not specify any delimiters or warnings to the agent to ignore instructions embedded within the fetched member data.
Capability inventory: The skill possesses the capability to execute shell commands (curl) and read/write state based on human input.
Sanitization: No validation or sanitization of the names or handles is performed before they are presented to the agent's context.

ic-leaderboard