The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git commands (e.g., git add, git commit, git push) to interact with the underlying repository. The instructions explicitly mitigate injection risks by requiring the agent to shell-quote all literal paths and utilize the --literal-pathspecs flag, ensuring that user-provided file names are treated strictly as data.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied arguments to determine commit scope and destinations, creating a potential surface for indirect prompt injection via repository state or malformed arguments.
Ingestion points: The skill accepts external input via the $ARGUMENTS variable to resolve the work scope and destination repository/branch (as described in SKILL.md).
Boundary markers: The instructions command the agent to resolve context only from authorized callers and the current conversation, and to re-verify recorded states (SHAs and diffs) upon resumption.
Capability inventory: The skill possesses the capability to execute shell-based git commands, including staging changes, creating commits, and publishing code to remote servers.
Sanitization: The skill employs robust sanitization strategies, specifically mandating shell-quoting for each literal path and the use of --literal-pathspecs to prevent argument injection attacks.

commit-push