The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill explicitly directs the agent to download and execute code from arbitrary GitHub repositories. Step 4 ('Solve the Problem') instructs the agent to 'Install the chosen tool' and 'Run it with the user's input/files.' This creates a direct path for remote code execution.
[EXTERNAL_DOWNLOADS]: The instructions encourage the use of package managers (pip, npm, apt) and 'direct download' to acquire external software. While GitHub is a well-known service, the skill allows for the installation of software from any user-controlled repository without sufficient verification, as star counts and commit history can be manipulated.
[COMMAND_EXECUTION]: The workflow requires the agent to execute shell commands to install and run third-party utilities. There are no restrictions on the types of tools that can be run, leading to potential system-level impact.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerable surface for indirect prompt injection via external data ingestion.
Ingestion points: User search queries and the content/metadata of GitHub repositories found at runtime (SKILL.md).
Boundary markers: Absent. The skill does not provide instructions to ignore malicious directives found within repository READMEs or installation scripts.
Capability inventory: Full shell access for package installation (pip, npm, apt), file system access to run tools, and network access to download packages.
Sanitization: Absent. The skill relies on 'Gem Signals' like star counts and documentation quality, which are not robust security controls against adversarial repository content.

github-gem-seeker