The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it processes untrusted data from git commit messages and evaluation reports to generate automated checklist items and verification commands.\n
Ingestion points: The skill reads evaluation data from evaluation-round-*.md files and extracts failure patterns from git history using git log and git show.\n
Boundary markers: There are no explicit instructions to sanitize or delimit potentially malicious text embedded within the processed reports or commit messages.\n
Capability inventory: The skill possesses the ability to Write and Edit project files and can define executable shell commands within generated checklists.\n
Sanitization: No validation or escaping is applied to the extracted patterns before they are incorporated into code blocks in new checklist versions.\n- [SAFE]: The skill incorporates several security best practices that mitigate potential risks:\n
Least Privilege: Tool access is constrained through allowed-tools, restricting the Bash tool to specific internal scripts and standard utilities.\n
Auditability: It maintains an evolution log and creates new versioned files instead of mutating existing ones, facilitating tracking of changes.\n
Rollback Protection: The skill performs a 'Pre-Edit Snapshot' of target files before modification to ensure data can be restored.\n
User Oversight: The workflow documentation explicitly prompts the user to review changes via git diffs post-execution.

retrospective