The Agent Skills Directory

[SAFE]: The skill implements a sophisticated project memory system designed to recall architectural decisions and conventions across sessions. The primary logic is contained in Node.js scripts (memory.mjs, mcp-server.mjs, memory-hook.mjs) that manage local Markdown files within a user-defined root directory (defaulting to ~/.ai-memory).
[SAFE]: Data exfiltration is mitigated through explicit security pipelines and 'audit' functions. The SENSITIVE_PATTERNS regex in memory.mjs actively looks for private keys, API tokens, and PII (emails/phones) before storage. It strictly follows a 'never persist secrets' policy.
[COMMAND_EXECUTION]: The install-memory-layer.mjs script executes commands such as claude mcp add or codex mcp add to register the MCP server. These are standard integration steps for agent platforms and are performed locally using node:child_process without downloading external payloads.
[EXTERNAL_DOWNLOADS]: The skill references several external memory providers (e.g., honcho, mem0, supermemory) and tools in the orchestrator-profile.md, but these are documented as 'optional additive' providers that require host-level configuration and user approval. The skill does not perform unauthorized network requests to these services.
[PRIVILEGE_ESCALATION]: The installer script provides a dry-run mode by default and only writes configurations to standard user-level application directories (e.g., AppData/Roaming, ~/Library/Application Support, ~/.config). It does not request sudo or modify system-level protected files.

memory-management