The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs various git operations (git status, git add, git commit, git apply) which are necessary for its documented purpose of managing commits. These operations are restricted to the local repository context.
[COMMAND_EXECUTION]: Uses a bundled Python script scripts/stage_hunks.py to perform non-interactive hunk staging. Technical review of the script confirms it uses subprocess.run() with argument lists rather than shell strings, which effectively mitigates shell injection risks.
[PROMPT_INJECTION]: The skill processes untrusted data from git diff and file contents. This represents a potential indirect prompt injection surface where a malicious file could attempt to influence the agent's behavior. However, the skill's capabilities are limited to staging and committing within the local repository, and it lacks network access to exfiltrate data.
Ingestion points: Untrusted data enters the agent context through git diff output in SKILL.md (Step 1) and the load function in scripts/stage_hunks.py.
Boundary markers: Absent. The skill does not use specific delimiters to wrap the diff content being analyzed.
Capability inventory: The skill can read repository state and modify the git index/commits. It cannot perform network operations, delete files outside the repository, or escalate privileges.
Sanitization: The Python script correctly uses subprocess argument arrays, which prevents input from being interpreted as shell commands.

commit-split