The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/gather_context.py executes local yadm commands (such as log, diff, and branch) to collect repository information. These commands are executed using subprocess.run with a list of arguments, which is the recommended practice for preventing shell injection attacks.
[PROMPT_INJECTION]: The skill processes data that may contain untrusted instructions, such as commit messages or content within file diffs, which represents an indirect prompt injection surface.
Ingestion points: Untrusted data enters the agent context via the output of yadm log and yadm diff within scripts/gather_context.py.
Boundary markers: The collected data is output as a raw JSON object without specific delimiters or "ignore" instructions to prevent the agent from following commands embedded in the commit history.
Capability inventory: The skill has access to execute yadm and python commands through the Bash tool as defined in the YAML frontmatter.
Sanitization: No sanitization or filtering is performed on the commit messages or diff contents before they are passed to the agent.

y-pr-desc