The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute standard shell commands including date, git log, and ghq list. These are used to collect metadata and history across local repositories for the purpose of identifying project patterns.
[DATA_EXPOSURE]: The skill is designed to perform broad searches across multiple repositories, including session history and recent activity. While this involves reading sensitive local data, the skill includes a specific safety rule ('Never leak secrets') prohibiting the inclusion of tokens, passwords, or API keys in the generated output.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection as it ingests untrusted data from git logs and repository files across the system.
Ingestion points: Reads data via git log, ghq, and a local dig.py script across project directories.
Boundary markers: None explicitly defined in the parallel agent instructions to delimit untrusted content.
Capability inventory: Performs file writes to the ψ/writing/dreams/ directory and executes shell subprocesses.
Sanitization: No specific sanitization or filtering of external content is mentioned before processing findings.

dream-original