The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow definition in workflows/smart_commit.toml constructs shell commands using string interpolation of tool arguments.
Input variables project_root and final_message are embedded into command strings like cd '{{project_root}}' and git commit -m "{{final_message}}" within the Qianji engine.
Lack of sanitization for shell metacharacters (e.g., quotes, semicolons) allows an attacker to execute arbitrary commands on the underlying system.
[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module to execute system commands across almost all operational scripts.
Scripts such as add.py, branch.py, commit.py, remote.py, and stash.py wrap git CLI calls, passing arguments directly from the AI agent's context to the shell.
[EXTERNAL_DOWNLOADS]: The smart_commit command executes cargo run, which is used to launch a local Rust-based engine.
This process may download, update, and compile external dependencies from registries like crates.io at runtime, bypassing static analysis of the skill's own code.
[COMMAND_EXECUTION]: The skill integrates with lefthook in scripts/prepare.py to run pre-commit hooks.
This mechanism allows the execution of arbitrary scripts defined within the repository being managed, which could lead to code execution if the repository contains malicious hook configurations.

git