The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute Node.js commands using node -e to interface with a local memory adapter (lib/memory/adapter.js). The instruction to pass data 'verbatim' into a shell command string represents a potential command injection surface if the agent does not properly escape the payload.
[DATA_EXFILTRATION]: The skill reads from local configuration files and state metadata located in the ~/.claude/ directory to determine the active memory backend and verify existing records before writing.
[PROMPT_INJECTION]: The skill acts as an ingestion surface for indirect prompt injection by persisting user-provided facts and preferences into long-term memory without explicit sanitization.
Ingestion points: Reads from ~/.claude/sustain/state.json, ~/.claude/CLAUDE.md, and the active conversation context.
Boundary markers: None present; stored memory content is treated as authoritative instructions or facts in future sessions.
Capability inventory: Uses the mcp__mempalace__store tool and shell execution via node -e to manage persistent state.
Sanitization: No sanitization, validation, or escaping of the memory content is performed prior to storage.

memory-write-router