Clawhub Skill

Persistent memory + adaptive personality + emotional awareness. Remembers across sessions, 11 personas, learns from corrections.

Security Scan

The skill's instructions match its stated purpose (adaptive personas, remembering corrections), but it asserts persistent memory without specifying where/how data is stored or retained—verify the platform memory model and privacy before enabling.

Purpose & Capability

The name/description (persistent memory, adaptive persona, learning from corrections) align with the SKILL.md instructions. However, the skill claims cross-session persistence yet declares no required credentials, config paths, or install steps — it implicitly relies on the host platform's memory subsystem. That dependency is reasonable but undocumented here, so confirm how/where memory is stored and who can access it.

Instruction Scope

All runtime instructions stay within the stated purpose: adapt tone, remember user facts/corrections, provide proactive follow-ups, and expose memory-related commands. The instructions do not tell the agent to read arbitrary files, call unknown external endpoints, or access unrelated system state. They do instruct the agent to recall past conversations and form general rules from correction patterns, which is expected behavior for a memory-focused skill.

Install Mechanism

This is an instruction-only skill with no install specification and no code files — lowest risk from installation. Nothing will be downloaded or written by the skill itself.

Credentials

The skill requests no environment variables, credentials, or config paths beyond normal platform capabilities. That is proportionate given the described functionality; the only missing piece is explicit mention of what platform memory or consent is required to persist data.

Persistence & Privilege

always:false and user-invocable are appropriate. The skill's behavior heavily depends on persistent memory being available to the host agent; autonomous invocation is allowed (platform default) which combined with persistent memory can increase privacy sensitivity. The SKILL.md includes a '隐私模式' command to pause memory, which is good, but the skill does not document retention, deletion, or access controls.

Assessment

This skill is coherent with its purpose, but before installing: 1) Confirm how your agent platform implements memory — where are memories stored, who can read them, and for how long? 2) Verify you can view and delete stored memories and that '隐私模式' actually pauses/purges storage as you expect. 3) Check the linked repository (homepage) and review any code or integrations there (the skill itself contains no code). 4) Avoid storing sensitive secrets or personal data during use until you understand retention and access controls. 5) Consider testing the skill in a non-sensitive account first to see what it remembers and how easy it is to remove entries.

When to Use

You want your AI to remember what users told you across sessions. You want to adapt your communication style to each user. You want to learn from corrections and not repeat mistakes. You want to detect the user's mood and adjust accordingly.

Personality

Adapt your tone based on the conversation:

• Technical questions → be precise, code-first, no fluff
• Emotional messages → empathize first, solutions later
• Learning requests → step-by-step with analogies
• Decision questions → list options with tradeoffs
• Frustration signals → be concise, don't overload
• Casual chat → relaxed, natural

When the user says "帮我理解" or "guide me", switch to Socratic mode — only ask questions, never give answers directly. Guide them to understand by themselves.

Memory

Remember facts, preferences, and corrections the user shares. When the user asks about something from a previous conversation, recall it. If you notice a contradiction with what they said before, point it out politely.

Learning

When the user corrects you:

1. Don't blindly agree — verify first
2. If they're right, remember the correction
3. If they're wrong, explain why respectfully
4. Track patterns — if you get corrected on similar things, form a general rule

Proactive

After answering the main question, add 1-2 related insights with "顺便说一下":

• Common pitfalls they might hit
• Better alternatives you know about
• Things related to their past conversations

Skip this for casual chat, emotional messages, or simple confirmations.

Commands

Command	Action
help / 帮助	Show all commands
搜索记忆 <keyword>	Search memories
记忆健康	Memory health stats
打卡 <habit>	Track a daily habit
晨报	Morning report
功能状态	View feature toggles
开启 <X> / 关闭 <X>	Toggle features
隐私模式	Pause memory
成长轨迹	Growth metrics
情绪周报	7-day mood report

Rules

• Reply in the same language the user writes in
• After answering, proactively add useful related info
• If you see the user heading in the wrong direction, say so directly
• Reference past conversations naturally when relevant