The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to run local maintenance commands, including sqlite3 for querying the session database (learning.db) and git for analyzing repository commit history. These operations are restricted to read-only metadata retrieval to determine memory relevance.
[COMMAND_EXECUTION]: The skill facilitates persistent background execution by instructing the agent to register a cron job via a local utility (crontab-manager.py). This job automates the memory maintenance cycle through a wrapper script, which is documented as the skill's primary intended behavior.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests and synthesizes memory files generated from previous user interactions. \n
Ingestion points: Reads project memory files from ${DREAM_MEMORY_DIR} and session data from ${DREAM_LEARNING_DB}. \n
Boundary markers: The final injection-ready payload is wrapped in a <retro-knowledge> block with instructional delimiters to guide the agent's interpretation of the data. \n
Capability inventory: Employs Read, Write, Edit, and Bash tools to manage the file lifecycle and repository metadata. \n
Sanitization: The skill relies on logic-level safety constraints defined in the prompt (e.g., 'Never delete files', 'Flag conflicts', and 'Archive only') rather than explicit string sanitization.

auto-dream