skill-evolution

Pass

Audited by Gen Agent Trust Hub on Apr 15, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements an "auto-evolution" mechanism that modifies skill templates based on user edit patterns. This creates a surface for Indirect Prompt Injection (Category 8). \n
  • Ingestion points: User edits are collected via PostTool hooks and stored in .claude/feedback/edit-patterns.jsonl (referenced in SKILL.md and pattern-detection-heuristics.md). \n
  • Boundary markers: Absent. The system matches regex patterns against raw user edits without delimiters or instructions to ignore embedded instructions. \n
  • Capability inventory: The skill uses Write and Edit tools to modify skill files and executes shell scripts in .claude/scripts/ (referenced in evolution-commands.md). \n
  • Sanitization: Absent. The logic focuses on pattern frequency and confidence rather than content sanitization before modifying skill instructions. \n- [COMMAND_EXECUTION]: The skill's implementation relies on external shell scripts (e.g., evolution-engine.sh, version-manager.sh) located in the project's hidden directory (.claude/scripts/). These scripts perform file operations and version management but are not provided in the skill bundle for security verification. \n- [COMMAND_EXECUTION]: Implementation subcommands interpolate shell variables like $SKILL_ID and $VERSION directly into command strings. If these values are derived from untrusted inputs or malformed filenames, they could potentially lead to command injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 15, 2026, 12:23 PM