The Agent Skills Directory

[PROMPT_INJECTION]: The skill incorporates specific defensive instructions to mitigate indirect prompt injection. It directs the agent to treat imported session history as evidence and not as current directives, preventing historical logs from overriding system instructions.
[COMMAND_EXECUTION]: The skill executes local Node.js scripts (read-session.mjs and read-text-chunk.mjs) to process session data. These scripts are self-contained, use built-in Node.js modules, and implement input sanitization for file path construction to prevent traversal attacks.
[DATA_EXFILTRATION]: The skill accesses the ~/.wheelmaker directory to retrieve session metadata and history. Access is restricted to reading local files for the purpose of summarization, with no network transmission or unauthorized data access observed.

handoff-wheel