memory-core-plus

Enhances OpenClaw memory by auto-recalling relevant past data before each turn and auto-capturing key info after each run for persistent context.

Security Scan

The plugin's code and instructions line up with its stated purpose: it auto-recalls relevant memories before turns and auto-captures durable facts after runs, writing them into workspace memory files — no unrelated credentials, external download URLs, or unexpected endpoints are requested.

Purpose & Capability

Name/description, SKILL.md, and source code are coherent: the plugin registers memory hooks, performs vector search via the memory search manager, and uses an LLM subagent to extract and persist memories. No unrelated environment variables or binaries are requested.

Instruction Scope

SKILL.md and the code precisely describe pre-turn recall and post-run capture. The capture path delegates writing to a subagent instructed to append to memory/YYYY-MM-DD.md — this gives the plugin (via the platform subagent) the ability to write workspace files, which is expected for a memory provider but worth noting. The SKILL.md includes examples of prompt-injection patterns only as blocked inputs (they are not being injected).

Install Mechanism

There is no remote download/install spec in the manifest; installation is via the OpenClaw plugin mechanism (openclaw plugins install). No arbitrary URLs, archive extraction, or shorteners are used.

Credentials

The skill declares no required environment variables, credentials, or config paths. Its runtime actions (search manager, subagent runs, writing memory files) are consistent with being a memory plugin and do not request unrelated secrets.

Persistence & Privilege

always:false (normal). The plugin registers persistent hooks (autoRecall/autoCapture) and uses a subagent to append to workspace memory files — this is appropriate for a memory provider but means it will persist user conversation content to disk unless configured otherwise.

Scan Findings in Context

[ignore-previous-instructions] expected: This pattern appears in the SKILL.md and safety code as a prompt-injection detection marker (the project filters such phrases before capture). Its presence is defensive, not an instruction to the LLM.

[you-are-now] expected: Used in the plugin's list of injection patterns to detect/jailbreak attempts; expected for a memory capture filter.

[system-prompt-override] expected: The SKILL.md and safety.ts reference common jailbreak phrases for detection. These scanner hits align with the plugin's stated safety checks.

Assessment

This plugin appears to do what it says: automatically surface past memories and append extracted facts to workspace memory files. Before installing, consider: (1) persistent storage — autoCapture appends conversation-derived facts to memory/YYYY-MM-DD.md, so private or sensitive data may be persisted; (2) subagent file writes — the plugin delegates extraction and file writes to a subagent (standard for memory providers) so confirm you trust the plugin source; (3) configuration options — you can disable autoCapture or autoRecall if you want manual control; (4) provenance — the package points to a GitHub repo but the registry source is 'unknown' here, so verify the upstream repo and its author if you need stronger assurance. If you are uncomfortable with automatic persistence, set autoCapture:false and/or review the code and repository before enabling.

Enhanced workspace memory with auto-recall and auto-capture for OpenClaw.

What It Does

memory-core-plus is an OpenClaw plugin that extends the built-in memory-core with two automated hooks:

• Auto-Recall — Before each LLM turn, semantically searches workspace memory and injects relevant memories into the prompt context.
• Auto-Capture — After each agent run, extracts durable facts, preferences, and decisions from the conversation and persists them to memory files.

Together they form a closed-loop memory system: information captured from past conversations is automatically surfaced when contextually relevant in future interactions.

Install

openclaw plugins install memory-core-plus
openclaw gateway restart

Auto-Recall and Auto-Capture are both enabled by default.

Configuration

{
  "plugins": {
    "entries": {
      "memory-core-plus": {
        "enabled": true,
        "config": {
          "autoRecall": true,
          "autoCapture": true
        }
      }
    },
    "slots": {
      "memory": "memory-core-plus"
    }
  }
}

Key	Type	Default	Description
autoRecall	boolean	true	Enable automatic memory recall before each agent turn
autoRecallMaxResults	number	5	Maximum memories to inject per turn
autoRecallMinPromptLength	number	5	Minimum prompt length (chars) to trigger recall
autoCapture	boolean	true	Enable automatic memory capture after each agent run
autoCaptureMaxMessages	number	10	Maximum recent messages to analyze for capture

Uninstall

openclaw plugins uninstall memory-core-plus
openclaw gateway restart

The gateway automatically falls back to the built-in memory-core.

Links

• GitHub
• npm