long-term-memory
ReviewAudited by ClawScan on May 10, 2026.
Overview
This is a coherent local long-term memory skill, but it deserves review because it persists and reinjects user context across sessions and its bundled tests can write unrelated sample memories into the real memory database.
Install only if you are comfortable with the agent keeping local long-term memories. Review what gets saved, avoid storing secrets or sensitive personal data, use project tags, and do not run the bundled tests against your real OpenClaw workspace unless they are changed to use a temporary database.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Information you share may be saved locally and later affect answers in future sessions, including across projects if not carefully tagged or filtered.
The skill explicitly asks the agent to persist user/project facts and reuse retrieved memories as future context. This is the core feature, but it means stored or stale information can influence later agent behavior.
1. On session start: Run `python3 scripts/memory_engine.py inject "current task"` and use the output as context 2. When user shares important info: Use `remember()` to save it 5. On session end: Summarize key outcomes for next session
Use this only if you want persistent memory. Avoid saving secrets, credentials, or sensitive personal data; add project tags; and ask the publisher to document review, deletion, retention, and exclusion controls.
Running the bundled tests could add someone else's facts, preferences, project details, or stock notes to your memory database, and the agent may later retrieve them as if they were your context.
The test uses the default persistent MemoryStore instead of a temporary test database, then writes hardcoded personal, project, and financial sample memories. If run, these records can pollute the user's real long-term memory.
store = MemoryStore()
store.remember("User: 王总. Workspace: preschool. Feishu communication.", ["user","preference"], 9, "manual", "core")
store.remember("恩华药业002262: 关注21.5买入, 董事长21.34增持, MACD金叉", ["stock","watch"], 7, "manual", "stocks")Do not run the tests against your real workspace. The publisher should change tests to use a temporary database, remove hardcoded personal/financial examples, and provide a cleanup command for demo/test memories.
Installing dependencies may fetch versions or model-related packages that differ over time.
Some dependencies are specified with lower-bound ranges rather than exact pins. This is common for Python projects, but it makes installs less reproducible and can pull newer dependency versions.
sentence-transformers>=2.2.0 numpy>=1.24.0 json5>=0.9.0
Install in a virtual environment, review package sources, and prefer a pinned lockfile before production use.