OpenClaw MongoDB Semantic Memory

MongoDB-backed long-term semantic memory for recalling, storing, searching, and managing facts, decisions, and user preferences across sessions.

This skill looks like a coherent long-term memory interface, but proceed cautiously. Key points to check before installing: - Ask the publisher for the source code or homepage and for details on the local daemon (http://localhost:7654): who runs it, what it stores, and where MongoDB/Voyage AI credentials are kept. - Verify ~/.openclaw/openclaw.json and any daemon config before enabling: the skill references this file but it was not declared in the registry metadata. - Be aware the skill has automatic hooks (auto-remember, session-to-memory, memory-bootstrap) that may persist information after every response. If you need tighter privacy, disable hooks (hooksEnabled: false) in the plugin config or prevent autonomous invocation. - Confirm how the daemon authenticates to MongoDB/Voyage AI and whether data ever leaves your environment (the SKILL.md doesn't explain credential flows). - If you install, test in an isolated environment and review stored memories for sensitive items; use memory_forget to remove data you don't want retained. Overall: the functionality matches a memory tool but lacking provenance, undeclared config usage, and automatic capture behavior make this suspicious until you get more operational detail.

Type: OpenClaw Skill Name: openclaw-memory-skill Version: 0.2.1 The `SKILL.md` documentation describes a `memory_get` tool that allows the agent to read specific files from its workspace using a `path` parameter. While the examples provided are benign (e.g., `MEMORY.md`), the lack of explicit path sanitization or restriction on the `path` parameter creates a significant information disclosure vulnerability. An attacker could potentially exploit this via prompt injection to instruct the agent to read sensitive files (e.g., configuration files, SSH keys) accessible within the agent's operating environment, even though there is no explicit instruction for exfiltration or other malicious actions within the skill itself.