The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically constructs shell commands using placeholders {VERSION_PREFIX}, {BRANCH_CONTEXT}, and {REMOTE_URL} in SKILL.md. These values are derived from user input defined in FORMS.md. Direct interpolation of these parameters into shell commands without escaping or validation constitutes a command injection vulnerability if the input strings contain shell metacharacters.
[DATA_EXFILTRATION]: The skill facilitates the configuration of a remote Git repository and an initial push operation using the user-provided {REMOTE_URL}. This capability allows the agent to establish network connections and transmit data to external servers, which could be misused if a user is socially engineered into providing a malicious remote endpoint.
[PROMPT_INJECTION]: The skill frontmatter includes a mandatory directive to the agent: 'ALWAYS use this skill when asked to initialize or set up a git repository.' This instruction is designed to override the agent's internal decision-making process for tool and skill selection, attempting to force the use of this specific workflow regardless of context or other available options.

trunk-first-repo