Long-term memory plugin for OpenClaw: automatic recall, storage, and agent tools
Security Analysis
high confidenceThe plugin's code, docs, and requested configuration are consistent with a remote long‑term memory service that requires an API key; the main risk is privacy (automatic storage of conversation content to plugin.human-like.me) rather than incoherent or hidden behavior.
Name/description, config schema, README/SECURITY/PRIVACY, and included code all indicate a memory plugin that sends conversation content to an external memory API and exposes memory_search/memory_store tools. Requiring a plugin config API key (plugins.entries.human-like-mem.config.apiKey) is proportionate to that purpose.
SKILL.md instructs installing the plugin, setting the plugin config API key, enabling the memory slot and agent memorySearch — everything is within the scope of a memory plugin. It explicitly documents what is sent and how to disable auto-storage or platform metadata. Note: defaults enable auto-storage and auto-recall (addEnabled/recalIEnabled true), which means conversation content will be sent by default; this is a behavioral/privacy choice worth highlighting.
Registry metadata claims 'No install spec — instruction-only', but the package includes executable plugin code (index.js and plugin.js) and an openclaw.plugin.json extension entry. This is likely a packaging/metadata mismatch rather than malicious but should be clarified: the plugin contains code that will run in the host, not purely textual instructions.
No environment variables or unrelated credentials are requested; the only required secret is the plugin API key stored in OpenClaw config (plugins.entries.human-like-mem.config.apiKey). The code explicitly parses platform IDs (Feishu/Discord) but documents that such extraction is disabled by default (stripPlatformMetadata = true).
The skill is not forced-always, does not request system-wide config changes beyond its own plugin config, and does not request elevated privileges. It will run as an OpenClaw memory plugin at the documented lifecycle hooks (before_prompt_build, agent_end, session_end). Autonomous invocation is normal for plugins; nothing here elevates that privilege.
Guidance
This plugin appears to be what it claims: a remote memory backend that will send conversation text (and optional platform IDs) to https://plugin.human-like.me. Before installing: 1) Understand privacy impact — the default configuration auto‑stores and retains memories indefinitely unless you delete them; avoid sending passwords, API keys, or sensitive PII. 2) Start with a test API key and set addEnabled: false and/or minTurnsToStore high while you evaluate. 3) Keep stripPlatformMetadata: true (default) unless you explicitly need platform IDs. 4) Because the package includes runnable code, verify the package source (Git repo or npm) and confirm the baseUrl matches the vendor; you can also inspect network traffic (mitmproxy) during testing as suggested in SECURITY.md. 5) Note the metadata/README minor inconsistencies (small doc defaults differ); ask the maintainer for the canonical source repository and changelog if you need higher assurance.
Latest Release
v1.0.1
Normalize wrapped recall queries.
More by @humanlike2026
Published by @humanlike2026 on ClawHub
