Memory Hamster
ReviewAudited by ClawScan on May 10, 2026.
Overview
Memory Hamster is a local memory-management skill, but it can persist and promote learned content into files that shape future agent behavior without a clear user-review gate.
Install only if you want a persistent local memory system for the agent. Before enabling cron or using promotion, back up the workspace, review what is stored in memory, avoid recording secrets, and require manual approval before anything is written into SOUL.md, AGENTS.md, TOOLS.md, or new skill files.
Findings (4)
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.
A mistaken or manipulated memory entry could change how the agent behaves, works, or uses tools in later sessions.
The skill instructs promoting learned content into persistent agent configuration/instruction files. The visible instructions do not require user review or provenance checks before promotion, so bad or untrusted learnings could persist into future sessions.
学习类型 | 提升到 ... 行为模式 | `SOUL.md` ... 工作流改进 | `AGENTS.md` ... 工具技巧 | `TOOLS.md` ... **添加** 到目标文件的相关部分
Require explicit user approval before modifying SOUL.md, AGENTS.md, or TOOLS.md; keep provenance for each promoted item; and maintain backups or an easy rollback path.
If enabled, the skill can archive memory files and create/update reflection records on a schedule without the user actively invoking it each time.
The documentation asks the user to configure scheduled jobs that run outside interactive sessions. This is disclosed and user-configured, but it creates continuing local activity after installation.
0 0 * * 0 /path/to/skills/memory-hamster/scripts/memory-gc.sh ... 45 23 * * * /path/to/skills/memory-hamster/scripts/nightly-reflection.sh ...
Only add the cron entries if you want ongoing background memory maintenance; use absolute paths, review logs, and remove the cron entries to disable the behavior.
Search results may surface older personal or project notes into the agent’s current context.
The search tool reads persistent personal learning and project-memory directories and prints previews. This matches the stated purpose, but those files may contain private information that can be reused across tasks.
const MEMORY_DIR = path.join(WORKSPACE, 'memory'); const LEARNINGS_DIR = path.join(WORKSPACE, '.learnings'); ... path.join(WORKSPACE, 'projects')
Do not store secrets in these memory files, and periodically review or prune memory entries you do not want reused.
Running the helper can add new skill scaffolds to the workspace, which may later influence agent capabilities or instructions if enabled or completed.
The helper can create a new skill directory and SKILL.md from user-provided input. It validates the skill name and relative output path, so this is purpose-aligned, but generated skills become persistent artifacts that should be reviewed.
mkdir -p "$SKILL_PATH" cat > "$SKILL_PATH/SKILL.md" << TEMPLATE
Use --dry-run first, inspect generated SKILL.md files before use, and do not automatically enable or publish generated skills without review.