The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by storing operational facts, session summaries, and decisions in a persistent store without explicit sanitization or boundary markers.
Ingestion points: Data enters the system through the memory_record_fact, memory_store_session, and memory_record_decision tools.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are defined for the stored content.
Capability inventory: The skill facilitates file system writes to ~/.openclaw/memory/ and interacts with other components like GAIT and HEARTBEAT, potentially allowing stored malicious instructions to affect these integrations.
Sanitization: The instructions do not mention any sanitization or validation of the input data before it is persisted.
[EXTERNAL_DOWNLOADS]: The skill documentation indicates that the first run will automatically download an embedding model (approximately 80MB) from an external source to enable semantic search capabilities.
[COMMAND_EXECUTION]: The troubleshooting section includes commands that execute external tools, specifically uvx --from netclaw-memory-mcp memory-mcp-server --help and sqlite3 for database verification.
[PERSISTENCE]: The skill is designed to maintain state across agent sessions by storing data in ~/.openclaw/memory/, which is a deliberate functional persistence mechanism.

memory