The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the repository's history and current changes to formulate its actions.
Ingestion points: The skill reads output from git status, git diff, and git log to summarize changes and determine the commit style.
Boundary markers: There are no instructions or delimiters to prevent the agent from following directives embedded in the analyzed code (e.g., instructions hidden in comments within a diff).
Capability inventory: The agent can execute git commands and formatting tools like ruff and prettier based on the context it gathers.
Sanitization: No sanitization or filtering is performed on the content read from the repository working tree or logs.
[COMMAND_EXECUTION]: The skill executes various git commands, including automated interactive staging with printf to split hunks. It also invokes standard development utilities such as ruff (via pixi) and prettier (via npx). While these are common tools, their execution is driven by the logic and context derived from the repository.

atomic-commits