The Agent Skills Directory

[Indirect Prompt Injection Consideration]: The skill processes data from external audit reports and finding descriptions, which constitutes a potential injection surface.
Ingestion points: Data enters the skill context through audit report files (--report), fresh output from tools like ruff or CodeQL, and specific finding IDs.
Boundary markers: The instructions include specific warnings to treat external content solely as data and not as commands, referencing established agentic guidelines for safety.
Capability inventory: The skill possesses capabilities for file system modification (applying fixes), git operations (committing changes), and shell command execution (python3, <audit-tool>).
Sanitization: The skill contains explicit instructions to detect, flag, and ignore instructions that may be embedded within the audit findings.
[Command Execution Pattern]: The skill executes system commands and local scripts as part of its verification and setup routine.
A pre-flight check executes python3 -m setup_preflight using a module located in the .apache-magpie-local directory.
All state-altering commands, such as git commit and gh pr create, are strictly controlled by 'Golden Rules' that require explicit human confirmation for every action.

audit-finding-fix