← Back to Skills Marketplace
346
Downloads
1
Stars
0
Active Installs
3
Versions
Install in OpenClaw
/install pafh-mini
Description
PAHF (Personalized Agents from Human Feedback) - Continual Personalization Framework. Triggered when applying the PAHF three-step loop: (1) Pre-action Clarif...
Usage Guidance
This skill appears coherent with its personalization purpose, but take these precautions before enabling it: (1) Verify and inspect existing files in ~/.openclaw/workspace/memory/ (MEMORY.md, USER.md, IDENTITY.md) and remove any sensitive secrets—the skill claims it will not store passwords/tokens but cannot remove pre-existing sensitive data. (2) Confirm you are comfortable with automatic 'daily observation' logging; if not, avoid enabling persistent storage. (3) Restrict file-system permissions on the memory directory so only you and the agent process can read/write it. (4) Test with a non-sensitive 'sandbox' identity before using real personal data. (5) Periodically audit the change logs (MEMORY.md and memory/YYYY-MM-DD.md) to verify what was recorded.
Capability Analysis
Type: OpenClaw Skill
Name: pafh-mini
Version: 1.0.2
The PAHF skill bundle implements a personalization framework for managing user preferences through local file storage (e.g., MEMORY.md, USER.md). It follows a structured three-step loop (clarification, action, and feedback) and includes explicit safeguards against storing sensitive information like passwords or API keys. The behavior is consistent with its stated purpose, and no indicators of data exfiltration, malicious execution, or unauthorized access were identified.
Capability Assessment
Purpose & Capability
The name/description (PAHF personalization) align with the declared behaviors: reading/writing local preference memory files and using memory_search/memory_get. No unrelated binaries, cloud creds, or external endpoints are requested.
Instruction Scope
SKILL.md explicitly instructs reading/writing MEMORY.md, memory/YYYY-MM-DD.md, USER.md, and IDENTITY.md under the local workspace and logging changes. This is within the personalization scope. Note: some actions (daily observations) are defined to be recorded automatically without confirmation — the skill requires overall consent for persistent storage but retains discretion about what counts as a daily observation vs. a core preference; users should confirm they are comfortable with that automatic logging behavior.
Install Mechanism
Instruction-only skill with no install spec or external downloads; lowest install risk. The declared tool dependencies (memory_search, memory_get) are reasonable as optional helpers and fall back to direct file reads.
Credentials
No environment variables, credentials, or config paths outside the local memory files are requested. Access is limited to local preference and identity files, which is proportionate to the stated purpose.
Persistence & Privilege
The skill persistently writes to local memory files and is user-invocable with autonomous invocation allowed (platform default). always:false and consent: true are appropriate. Users should note that the skill is allowed to log 'daily observations' without per-event confirmation and will create/modify files under ~/.openclaw/workspace/memory/.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install pafh-mini - After installation, invoke the skill by name or use
/pafh-mini - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.2
PAHF-mini v1.0.2
- Added explicit privacy, consent, and write-confirmation policies to the skill manifest and documentation.
- Declared formal dependencies on `memory_search`, `memory_get`, and required memory file read/write permissions.
- Defined new writable memory locations, including user-specific preference files.
- Updated documentation to detail when and how user confirmation is requested before recording preferences.
- Outlined sensitive data handling: explicitly states that such data is never stored.
- No code changes; all enhancements are in metadata and documentation for clarity and consent management.
v1.0.1
**PAHF-mini v1.0.1 Changelog**
- Added a new reference document: `references/preference-schema.md` to clarify preference memory structure.
- Updated core framework description for broader use cases and clarity.
- Expanded details on memory file structure, preference retrieval, and change tracking.
- Enhanced examples for pre-action clarification, ambiguity handling, and preference drift adaptation.
- Emphasized best practices and highlighted integration with existing memory systems.
v1.0.0
PAHF Mini 1.0.0 初始发布
- 推出精简版 PAHF，实现个性化用户记忆管理
- 支持行动前澄清、行动后修正、偏好漂移自动检测
- 明确澄清条件与反馈处理，加入精简版用户记忆模板
- 内置与其他相关 skill 的协作说明
- 提供操作与实现检查清单方便快速接入
Metadata
Frequently Asked Questions
What is PAFH-Memory?
PAHF (Personalized Agents from Human Feedback) - Continual Personalization Framework. Triggered when applying the PAHF three-step loop: (1) Pre-action Clarif... It is an AI Agent Skill for Claude Code / OpenClaw, with 346 downloads so far.
How do I install PAFH-Memory?
Run "/install pafh-mini" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is PAFH-Memory free?
Yes, PAFH-Memory is completely free (open-source). You can download, install and use it at no cost.
Which platforms does PAFH-Memory support?
PAFH-Memory is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created PAFH-Memory?
It is built and maintained by Wenju Gao (@welkeyever); the current version is v1.0.2.
More Skills