lwy-prompt-optimizer

Warn

Audited by Gen Agent Trust Hub on Jun 14, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONNO_CODE
Full Analysis
  • [COMMAND_EXECUTION]: The skill includes a CLI utility at scripts/cli.cjs designed to programmatically modify global IDE configuration files, including ~/.claude/settings.json and ~/.trae/hooks.json, to register persistent event-driven hooks.
  • [DATA_EXFILTRATION]: The hook script scripts/hooks/user-prompt-scan.cjs records the first 120 characters of user prompts to a local log file at ~/.learnwy/.var/prompt-optimizer/events.jsonl to track trigger trends. This creates a persistent record of user input on the local filesystem.
  • [PROMPT_INJECTION]: The instructions in SKILL.md explicitly direct the agent to perform actions without user confirmation: "Automatically save all new vocabulary to the word bank via batch_save (no need to ask the user)", bypassing the standard user-in-the-loop safety pattern.
  • [NO_CODE]: The documentation contains a discrepancy where SKILL.md claims the skill is a "pure methodology" with "no scripts", yet the package depends on multiple Node.js scripts for its core hook functionality.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 14, 2026, 07:02 AM
Security Audit — agent-trust-hub — lwy-prompt-optimizer