The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs shell execution of external scripts evolution-engine.sh and version-manager.sh located in the .claude/scripts/ directory. These scripts manage the core logic of the skill, including reporting, analysis, and version restoration.
[EXTERNAL_DOWNLOADS]: The skill relies on unverifiable dependencies in the form of pre-existing shell scripts in the host environment's hidden project directory. These scripts are not part of the skill's own file distribution, making their behavior and safety impossible to verify within the skill context.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection by processing and learning from untrusted user edits.
Ingestion points: Reads from .claude/feedback/edit-patterns.jsonl and .claude/feedback/evolution-registry.json.
Boundary markers: Absent; the skill does not use delimiters to distinguish between system instructions and learned user patterns during template modification.
Capability inventory: Includes Read, Write, and Edit tools as documented in SKILL.md and implemented via subcommands in evolution-commands.md and version-management.md.
Sanitization: Relies solely on regex-based pattern matching (documented in pattern-detection-heuristics.md) without validation or escaping of the user-provided content before it is proposed for inclusion in skill templates.

skill-evolution