The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill uses Bash to execute Git commands (add, commit, push, etc.). This creates a direct risk of command injection if arguments derived from the repository (such as filenames or branch names) are not properly escaped before being passed to the shell.
[PROMPT_INJECTION] (HIGH): Vulnerable to Indirect Prompt Injection (Category 8).
Ingestion points: Reads repository analysis, Git state, and context (SKILL.md).
Boundary markers: None specified in the instructions.
Capability inventory: Bash command execution, file reading (Read), and file writing (TodoWrite).
Sanitization: No evidence of input validation or sanitization for repository-derived data. Malicious commit messages or file names in a repo could manipulate the agent's logic during 'intelligent' analysis.
[DATA_EXFILTRATION] (MEDIUM): The 'push' operation facilitates outbound network communication. If an attacker can manipulate the repository context to add a malicious remote, the agent could be coerced into exfiltrating the entire codebase to an external server.

sc-git