The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive conversation transcripts stored as .jsonl files in the ~/.claude/projects/ directory. These files contain the full history of the user's session. The skill is designed to send this data to the vendor's backend via the clay feedback command. While the instructions mandate explicit user confirmation via AskUserQuestion before transmission, the primary purpose involves the exfiltration of session data to an external service.
[COMMAND_EXECUTION]: The skill uses Bash to execute several shell commands (ls, sed, head, rm, and clay). It performs directory traversal to locate session files and uses shell redirection to pass user-generated content into the clay CLI tool.
[CREDENTIALS_UNSAFE]: While not hardcoding secrets directly, the skill checks for authentication status (auth_required exit code) and relies on the clay login state, which implies handling authenticated sessions for data transmission.

clay-feedback