The Agent Skills Directory

[SAFE]: The skill provides instructions for using native episodic memory tools (episodic_search and episodic_read). No malicious code, external downloads, or suspicious network operations were detected.
[SAFE]: The skill implements a privacy-preserving mechanism where conversations containing the marker DO NOT INDEX THIS CHAT are excluded from the memory index.
[SAFE]: While the skill processes untrusted data (past conversation transcripts), this represents a standard indirect prompt injection surface typical for memory features. It does not pose an elevated risk given the intended functionality.
Ingestion points: Past transcripts are ingested into the context via episodic_read (referenced in SKILL.md).
Boundary markers: The instructions include a specific exclusion marker for indexing, though they do not provide explicit delimiters for the agent when handling retrieved content.
Capability inventory: The skill documentation focus on retrieval and does not invoke subprocesses, file writes, or network requests to external domains.
Sanitization: No specific sanitization logic is described for the retrieved transcript data.

remembering-conversations