The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local shell scripts (evolution-engine.sh and version-manager.sh) located in the project's .claude/scripts/ directory to perform reporting, analysis, and version rollback operations as documented in evolution-commands.md.
[PROMPT_INJECTION]: The skill implements an Auto-Evolution system that analyzes user-performed edits from .claude/feedback/edit-patterns.jsonl to suggest modifications to skill instructions. This presents an indirect prompt injection risk where adversarial user input could be incorporated into the agent's core skill set.
Ingestion points: Raw user edits stored in .claude/feedback/edit-patterns.jsonl.
Boundary markers: The system uses a human-in-the-loop review process via AskUserQuestion before applying changes, but lacks structural delimiters or automated sanitization for the ingested content.
Capability inventory: Possesses the ability to modify other skills using Write and Edit tools and execute administrative shell scripts.
Sanitization: Relies on regex matching for categorization, which does not sanitize the underlying content of the edit pattern being suggested for inclusion.

skill-evolution