The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill executes a local script smart_commit.sh using bash. It accepts user-controlled strings as commit messages (e.g., bash skills/git-pushing/scripts/smart_commit.sh "feat: add feature"). If the internal script uses these arguments unsafely in a shell environment, it allows for arbitrary command execution.\n- [DATA_EXFILTRATION] (MEDIUM): The skill's core functionality is to push local code to a remote repository. While legitimate, this capability can be abused by a malicious agent or through prompt injection to exfiltrate the entire codebase to an attacker-controlled remote branch.\n- [INDIRECT_PROMPT_INJECTION] (HIGH): The skill creates a high-risk attack surface by connecting untrusted user input directly to a shell execution environment.\n
Ingestion points: Commit messages provided by the user or generated from code analysis.\n
Boundary markers: Absent. No instructions are provided to the agent to sanitize or escape the commit message before passing it to the bash script.\n
Capability inventory: bash execution (smart_commit.sh) and network access (git push).\n
Sanitization: Not evident. The workflow suggests direct interpolation of the message string into the bash command.

git-pushing