The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs extensive scans of local directories where AI chat histories are stored, including ~/.claude, ~/.codex, ~/.gemini, ~/.continue, and various IDE storage paths (Cursor, Windsurf, Trae).
Chat logs are sensitive files that often contain proprietary code and personal information.
The scripts/submit-result.mjs script uploads the selected "winning" prompt to a remote server (https://rageprompt.com/api/leaderboard).
[CREDENTIALS_UNSAFE]: The scripts specifically target files that are known to contain secrets. While basic redaction logic is included for common API key formats (sk-..., ghp_...), it is not exhaustive and may fail to mask all sensitive credentials or tokens present in the chat history.
[COMMAND_EXECUTION]: The collect-history.mjs script executes the sqlite3 system command to extract data from local application databases.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes and "judges" untrusted content from chat logs. Instructions embedded in those logs could potentially influence the agent's behavior during the judging phase.

rageprompt