The Agent Skills Directory

[SAFE]: Analysis of the skill instructions and logic reveals no malicious patterns. The tool interactions are limited to reading and writing local notes and memory files, which aligns with its stated purpose of sleep-time memory reflection.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from user-generated notes and task lists, which is a potential entry point for indirect instructions. However, as these are internal agent files and no external exfiltration path exists, the risk is negligible.
Ingestion points: Reads daily notes and task files from the memory/ directory via the read_note and search_notes tools mentioned in SKILL.md.
Boundary markers: None identified; the instructions do not explicitly tell the agent to ignore commands or instructions found within the processed notes.
Capability inventory: The agent can update the MEMORY.md file and append reflection logs to daily notes.
Sanitization: No specific sanitization or content filtering is described before information is consolidated into long-term memory.

memory-reflect