The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from pull request titles, bodies, and diffs, creating a surface for indirect prompt injection. Malicious instructions embedded in a pull request could attempt to influence the agent's logic during the analysis or proposal phases.
Ingestion points: Pull request metadata (title, body) and diff content are retrieved via gh pr view and gh pr diff commands in SKILL.md.
Boundary markers: No explicit delimiters or instructions are provided to the agent to isolate untrusted PR data from its core workflow logic.
Capability inventory: The skill is capable of modifying files via git apply, pushing code to remote branches, and creating new pull requests via the gh CLI.
Sanitization: No sanitization, validation, or filtering of the fetched pull request content is performed.
[COMMAND_EXECUTION]: The skill utilizes the gh and git command-line tools to interact with the NVIDIA/Megatron-LM repository. These operations include analysis of existing pull requests and the creation of new draft pull requests on the user's fork.

mcore-split-pr