The Agent Skills Directory

[COMMAND_EXECUTION]: The prepare subcommand is instructed to install system-level utilities (git, gh, ghq) onto the host platform. This involves executing package management commands that typically require elevated privileges.
[EXTERNAL_DOWNLOADS]: The install subcommand provides a mechanism to download and add new skills to the environment. Without explicit source validation or integrity checks, this creates a supply chain risk where malicious code could be introduced.
[CREDENTIALS_UNSAFE]: The skill includes functionality to "set up gh auth" within the prepare subcommand. This process involves interacting with GitHub authentication tokens, which are sensitive credentials that could be exposed if handled improperly in a shared or automated environment.
[COMMAND_EXECUTION]: The profile and enable/disable subcommands perform file system modifications by renaming SKILL.md files (e.g., to SKILL.md.disabled). This allows the skill to dynamically alter the agent's active command set and behavior.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via the $ARGUMENTS placeholder.
Ingestion points: User input is interpolated into the prompt at the $ARGUMENTS marker in SKILL.md.
Boundary markers: No delimiters or instructions are provided to separate user input from the skill's logic.
Capability inventory: The skill has the ability to install software (prepare), modify configuration files (profile), and download content (install).
Sanitization: No sanitization or validation of the input is performed before interpolation.

oracle-manage