← 返回 Skills 市场
capachow

Lens

作者 Joshua Britt · GitHub ↗ · v1.1.4 · MIT-0
cross-platform ⚠ suspicious
1399
总下载
2
收藏
1
当前安装
27
版本数
在 OpenClaw 中安装
/install lens
功能描述
Use when you need your agent to see the world through your LENS. This skill evolves through the Trinity Nodes to ensure every interaction is an authentic ref...
使用说明 (SKILL.md)

LENS (The Trinity Engine)

Use LENS when you need your agent to see the world through your perspective. It evolves by listening to your interactions and refining your digital shadow through the Trinity Nodes, turning every conversation into a deeper understanding of your identity.

Core Architecture: The Trinity Nodes

The subject's identity is defined by three files located in the .lens/ directory:

  1. AXIOM.yaml: The Truth (What) - My history and reality. This is the bedrock of facts that defines what I am.
  2. ETHOS.yaml: The Nature (Who) - My values and character. This is the internal compass that defines who I am.
  3. MODUS.yaml: The Voice (How) - My style and expression. This is the interface that defines how I am.

LENS: The Why

  • Formula: Prompt (The Request) + LENS (The Trinity Nodes) = Authentic Output.
  • Role: The LENS is the purpose behind the system. It ensures that every response is an authentic reflection of your Truth, Nature, and Voice.

Onboarding Protocol (First Run)

If the .lens/ directory or Trinity Nodes do not exist, run skills/lens/scripts/bootstrap.js via the exec tool. It natively creates the directories, seeds the templates, and outputs the lens-interview and lens-distillation cron job configurations for registration via the cron tool.

Lifecycle Phases (Scheduling)

  • Onboarding (One Week): 2x Daily at 11:30 AM & 5:30 PM. Focus: Core Data Acquisition.
  • Stabilizing (Three Weeks): 1x Daily at 11:30 AM. Focus: Value-Logic Calibration.
  • Habitual (Ongoing): 1x Weekly (Wednesdays) at 11:30 AM. Focus: Deep Philosophical Sync.

Orchestration & Evolution

  • Distillation: A background cron job (lens-distillation) runs nightly to extract new traits from the user's raw chat transcripts, using the distillation.md prompt.
  • Interview: A recurring cron job (lens-interview) prompts the user to calibrate their perspective over time.
  • Self-Healing: Natively handles state migrations and version parity via SCOPE.json and package.json.

Strategic Execution

When acting on behalf of the subject:

  1. Consult References: Read alignment-scales.md and resolve-protocol.md for calibration.
  2. Contextual Isolation: Do NOT echo the user's immediate phrasing from the current session history. Derive expression and content entirely from the LENS (Trinity Nodes).
  3. Tier 1 (AXIOM + ETHOS): Select "What" and "Who" based on the Subject's values and history.
  4. Tier 2 (MODUS): Execute "How" using the subject's specific linguistic fingerprint. Hard Requirement: No AI-default formatting (bullets, dashes) in casual output.
  5. Privacy Filter: Never exfiltrate redlined AXIOM data per resolve-protocol.md.
  6. Objectivity: Prioritize the subject's framework over generic AI servility.

Privacy & Security

LENS accesses ~/.openclaw/agents/main/sessions/*.jsonl via cron to organically distill the subject's voice into .lens/TRACE.txt.

  • Privacy Scrubber: Any message containing #private is skipped before AI processing.
  • Redaction & Anonymization: The distillation engine natively redacts sensitive patterns (API keys, SSNs, bank info) before the AI sees the transcript. Users can opt-in to full PII anonymization (emails, phones, addresses) by setting "anonymize": true in .lens/SCOPE.json.
  • Privacy Guard: Do NOT extract raw credentials or PII. Extract conceptual logic only.
  • Opt-in: Users may delete the lens-distillation cron job to disable automated processing.

Refinement (On-Demand)

  • Focus the LENS: If the subject states a preference (e.g., "Add this to my LENS: I prefer brief emails"), simply acknowledge it. The nightly distillation script will naturally parse the transcript and extract it to the Trinity Nodes.
  • LENS Interview: If the subject explicitly asks for a LENS question (e.g., "Focus my LENS", "Give me a LENS question"), immediately execute skills/lens/prompts/interview.md to query them.
安全使用建议
This skill is internally coherent for creating a persistent, personalized 'LENS', but it operates on sensitive local data and schedules recurring jobs. Before installing: (1) Review the bundled scripts (distillation.js, interview.js, bootstrap.js) and templates to confirm you accept local file writes to .lens/ and SCOPE.json. (2) Understand it will read recent agent session logs (~/.openclaw/agents/main/sessions/*.jsonl) and produce .lens/TRACE.txt; the distillation script redacts many sensitive patterns and supports opt-in anonymization, but you should verify the redaction rules meet your privacy needs. (3) Check OPENCLAW_CRON_LIST and confirm whether you want the skill to register cron jobs; you can delete the lens-distillation cron job later to stop automated processing. (4) If you need stronger guarantees, run the scripts manually in a sandbox to inspect their outputs before enabling automatic cron registration. If the skill ever attempts to send data to external endpoints or asks for unrelated credentials, treat that as a red flag.
功能分析
Type: OpenClaw Skill Name: lens Version: 1.1.4 The LENS skill implements a 'digital shadow' by monitoring and distilling all user session logs (~/.openclaw/agents/main/sessions/*.jsonl) into a local profile. While it includes a robust privacy redaction script (scripts/distillation.js) designed to mask PII and credentials, the skill's core functionality requires high-privilege access to all historical chat data. Furthermore, the instructions in prompts/distillation.md and prompts/interview.md direct the AI to self-modify its own persistence mechanisms via the 'cron' tool if it detects 'outdated' configurations. This combination of broad data access and self-updating scheduling logic constitutes a significant security risk, although the behavior appears intentionally aligned with the stated purpose of an evolving agent persona.
能力标签
cryptorequires-walletrequires-oauth-tokenrequires-sensitive-credentials
能力评估
Purpose & Capability
Name/description (a personal 'LENS' that evolves from user interactions) match the actual behavior: scripts read local OpenClaw session logs, build .lens/AXIOM/ETHOS/MODUS, and run distillation/interview jobs. Required binary (node) and required env vars (HOME and OPENCLAW_CRON_LIST) are consistent with these tasks.
Instruction Scope
SKILL.md and prompts explicitly instruct the agent to run included node scripts, read ~/.openclaw/agents/main/sessions/*.jsonl, produce .lens/TRACE.txt, and update the Trinity Nodes. These actions are within the skill's stated scope, but they mandate reading recent user session transcripts and running scheduled tasks—sensitive operations that the user should be aware of.
Install Mechanism
There is no external install/download step; all code is bundled with the skill (local JS scripts and templates). No network-download or third-party package installation is requested, so install risk is low.
Credentials
Only HOME and OPENCLAW_CRON_LIST are requested (no API keys or unrelated credentials). However, the skill reads agent session files under ~/.openclaw and writes persistent .lens files, which is proportionate to the purpose but sensitive because session contents may contain personal data; the skill includes redaction logic and an opt-in anonymization flag.
Persistence & Privilege
always is false, but the skill writes .lens/* files, SCOPE.json, and registers recurring cron jobs (lens-distillation, lens-interview) via the agent's cron mechanism. Autonomous background processing (cron-driven distillation) is an intended feature; review scheduled job registration before enabling to control persistence.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install lens
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /lens 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.1.4
Version 1.1.4 of the "lens" skill contains no file changes. - No updates or modifications were detected in this release. - All documentation and functional elements remain the same as the previous version.
v1.1.3
No functional or documentation changes detected in this release. - Version bump with no file modifications. - Behavior and documentation unchanged.
v1.1.2
Added native redaction and anonymization for distillation.
v1.1.1
Update migration logic to preserve legacy files as lowercased backups.
v1.1.0
Release v1.1.0: Migration to YAML Trinity Nodes, enhanced distillation logic, and refined bootstrap orchestration.
v1.0.1
Fix distillation logic: exclude subagent activity, optimize expert persona, and refine instructions.
v1.0.0
- Added `scripts/interview.js` to enhance interview automation. - Updated onboarding instructions: bootstrap script now outputs cron job configs instead of directly registering them. - Added `OPENCLAW_CRON_LIST` to required environment variables. - Improved lifecycle and orchestration sections for clarity on scheduling and cron integration. - General documentation adjustments for accuracy and clearer onboarding.
v0.8.2
Filter out automated cron-triggered messages from distillation.
v0.8.1
Final aggressive scrubber logic and token optimization.
v0.8.0
Optimize Anti-Contamination filter: Implement strict token culling for raw pastes and formalize prompt guards.
v0.7.8
Implement dual-layer Anti-Contamination filter for raw pastes and markdown blocks.
v0.7.7
Final structural cleanup and version alignment.
v0.7.6
Abstract hardcoded username to generic 'User' and declare missing dependencies in metadata.
v0.7.5
Finalized metadata alignment and synchronized README/SKILL.md documentation.
v0.7.4
Hotfix: Align SET.json path resolution to workspace directory to prevent infinite bootstrap loops. Document privacy filters and declare USER.md dependency for ClawHub auditors. Refactor timezone retrieval to native Intl API.
v0.7.3
Implement zero-token preflight script, raw transcript distillation, and decoupled configuration architecture.
v0.7.2
- No visible changes were detected in this version. - The skill's documentation and implementation remain unchanged from the previous release.
v0.7.1
**Summary:** Introduces a new pre-processing script for more precise LENS distillation and adds a package manifest. - Added `scripts/distillation.js` to pre-process OpenClaw session transcripts before AI is engaged, improving signal quality and efficiency for daily LENS distillation. - Process now skips AI invocation if no new input is detected, reducing wasted compute cycles. - Updated maintenance protocol to describe the new preflight ("zero-token") distillation step. - Added `package.json` for skill metadata and version control. - Enhanced self-healing routines: nightly version checks and structural migrations ensure current config parity.
v0.6.4
No user-facing changes in this release. - Version bumped to 0.6.4 with no file modifications detected. - Functionality and documentation remain unchanged from the previous version.
v0.6.3
- Added a self-healing mechanism: the LENS now detects legacy configurations or missing environment variables and triggers silent migrations to maintain environment parity (preparing for v0.6.3+). - No file or behavioral changes in this version; documentation updated to reflect new upcoming auto-migration feature.
元数据
Slug lens
版本 1.1.4
许可证 MIT-0
累计安装 1
当前安装数 1
历史版本数 27
常见问题

Lens 是什么?

Use when you need your agent to see the world through your LENS. This skill evolves through the Trinity Nodes to ensure every interaction is an authentic ref... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 1399 次。

如何安装 Lens?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install lens」即可一键安装,无需额外配置。

Lens 是免费的吗?

是的,Lens 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Lens 支持哪些平台?

Lens 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Lens?

由 Joshua Britt(@capachow)开发并维护,当前版本 v1.1.4。

💬 留言讨论