The Agent Skills Directory

[SAFE]: The skill implements a structured multi-step workflow (1 through 6) designed to process data in memory and present a consolidation report for user review before any modifications are applied.
[COMMAND_EXECUTION]: Executes a local Python script (parse_mem0_config.py) to retrieve retention policies from the project's configuration. This script is part of the skill's own package and is used for platform-specific path resolution.
[SAFE]: In automated mode, the skill uses a file-based lock mechanism at /tmp/mem0_dream_auto.lock to ensure process synchronization and prevent concurrent modifications of the memory store.
[SAFE]: Memory management operations (fetching, merging, and pruning) are scoped strictly to the current user_id and app_id, utilizing internal tool calls rather than external network requests.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data (user memories) which represents a potential attack surface.
Ingestion points: Memory content is retrieved via the get_memories tool in Step 2.
Boundary markers: The skill does not explicitly use delimiters for memory content, but instructions require the agent to analyze content heuristically (similarity/noun overlap).
Capability inventory: The skill has the ability to delete (delete_memory) and create (add_memory) entries based on the processed data.
Sanitization: The skill mitigates risks by requiring user confirmation of a structured diff report (Step 4) before execution, and in auto-mode, it restricts high-risk operations like contradiction resolution to manual review.

dream