The Agent Skills Directory

Dynamic Script Generation and Execution: The skill is instructed to write a script (such as Python or Bash) on-the-fly to the workspace and subsequently execute it to process repository logs. Generating and running executable code at runtime is a security consideration for system integrity and requires that the agent's logic for script creation remains consistent with safety policies.
Indirect Prompt Injection Surface: The skill ingests commit messages, titles, and diffs from the repository's history (VCS) and sends this data to a language model for analysis. Because version control history is often composed of input from multiple contributors and can contain arbitrary text, it represents an ingestion point for untrusted data that could potentially attempt to influence model behavior during processing.
Command Execution on Live Repository: The instructions specify that version control commands (such as git log, diff, or blame) must be executed within the live repository root to access metadata. This capability allows the skill to interact directly with the project's historical data and metadata, which is necessary for its stated goal but constitutes a broad level of access to the codebase's history.

mantis-history