The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell scripts to manage a local session store in /tmp. The script includes security hardening such as symlink verification, ownership checks, and restrictive permissions (chmod 700).
[PROMPT_INJECTION]: The resume command creates an indirect prompt injection surface by ingesting external data. However, the skill provides specific defensive instructions: 1. Ingestion points: Artifacts, URLs, or folder contents supplied by the user via the resume route. 2. Boundary markers: The skill explicitly mandates treating external content as untrusted context and not instructions. 3. Capability inventory: The skill utilizes local shell execution for storage setup and file read/write for snapshots. 4. Sanitization: The creation process includes an explicit instruction to redact secrets, credentials, and unrelated personal information.

ce-handoff