← 返回 Skills 市场
Intent-Engineering
作者
Daniel Foo Jun Wei
· GitHub ↗
· v1.0.0
· MIT-0
79
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install intent-engineer
功能描述
A meta-framework for designing, building, and orchestrating an ecosystem of strategically-aligned agent skills. This skill governs how the agent itself opera...
使用说明 (SKILL.md)
Intent Engineering: The Agent's Operating System
Overview
This skill is more than a tool; it is the operating system for the agent itself. It provides a comprehensive meta-framework for building and managing an ecosystem of interconnected agent skills. When you ask the agent to build something, it uses this very skill to guide its own reasoning, decision-making, and implementation process.
This framework extends the principles of intent engineering to a multi-skill environment and, critically, to the agent's own behavior. It introduces a structured approach to ecosystem architecture, data governance, skill composition, shared intent, and agent self-governance.
The Agent as the Orchestrator
The agent is not just a passive tool; it is the active orchestrator of this entire framework. This creates a virtuous cycle of recursive improvement.
- Intent Amplification: The agent takes your high-level, sometimes "shallow," prompts and uses this framework to translate them into well-architected, robust, and aligned skills.
- Complexity Absorption: The agent handles the intricate details of data contracts, orchestration patterns, and governance, allowing you to focus on strategic intent.
- Self-Referential Governance: The agent applies the principles of this framework to itself. Its decisions are logged, its outputs are validated against data contracts, and its actions are aligned with the shared intent. This is meta-governance.
- Recursive Improvement: The agent uses the
intent-engineeringskill to improve and extend theintent-engineeringskill itself, creating a self-improving system.
The Intent-Driven Skill Ecosystem Architecture
An aligned skill ecosystem consists of five core components that work together to ensure that individual skills are greater than the sum of their parts.
| Component | Description | Implementation |
|---|---|---|
| 1. Skill Registry | A centralized, machine-readable inventory of all available skills, their capabilities, dependencies, and data contracts. | references/skill_registry.json |
| 2. Data Contracts | Formal schemas (JSON Schema) defining the inputs and outputs for each skill, ensuring predictable and reliable data exchange. | references/data_contracts/ |
| 3. Orchestration Engine | A system for defining and executing workflows that compose multiple skills, handling data flow, and managing dependencies. | scripts/orchestrator.py |
| 4. Shared Intent Framework | A global set of organizational goals, values, and decision boundaries that all skills inherit, ensuring consistent alignment. | references/shared_intent.md |
| 5. Agent Decision Framework | The internal guidance system the agent uses to apply this framework, amplify user intent, and govern its own actions. | references/agent_decision_framework.md |
The Enhanced 4-Phase Workflow
The agent follows this workflow when you ask it to build or modify a skill.
Phase 1: Deconstruct Intent (Ecosystem-Aware)
Objective: To define a skill's strategic purpose within the context of the broader ecosystem.
New Workflow Steps:
- Define Skill's Role: In addition to its own goal, define how this skill contributes to the overall ecosystem.
- Align with Shared Intent: Consult the
references/shared_intent.mdto ensure the skill's values and boundaries are consistent with organizational-level principles. - Identify Dependencies: Use the
references/skill_registry.jsonto identify any existing skills this new skill will depend on.
Phase 2: Map Capabilities & Define Data Contracts
Objective: To define the skill's tasks and formalize its data interfaces.
New Workflow Steps:
- Design Workflow: Decompose the skill's tasks as before.
- Define Data Contracts: For each input and output, create a formal JSON Schema in the
references/data_contracts/directory. - Specify Data Lineage: Document where the skill's input data comes from and where its output data goes.
Phase 3: Build Infrastructure & Register the Skill
Objective: To build the skill's resources and make it discoverable by the ecosystem.
New Workflow Steps:
- Build Resources: Create scripts and templates as before.
- Register the Skill: Add a new entry to the
references/skill_registry.json.
Phase 4: Implement, Orchestrate, and Iterate
Objective: To implement the skill's logic, including its interactions with other skills.
New Workflow Steps:
- Implement Logic: Write the core logic for the skill.
- Orchestrate Interactions: Use the
scripts/orchestrator.pyto call other skills. - Validate and Deliver: Validate the skill and its interactions within the ecosystem.
Resources for Ecosystem Orchestration
This skill now includes a richer set of resources to manage the entire ecosystem:
references/shared_intent.md: Defines the global values and goals for the entire organization.references/skill_registry.json: A central catalog of all skills.references/data_contracts/: A directory containing all data contract schemas.references/agent_decision_framework.md: The agent's internal guidance for applying this framework.references/recursive_skill_development.md: A guide on how the agent can improve this skill itself.scripts/orchestrator.py: A Python script for composing and executing multi-skill workflows.templates/agent_audit_log.md: A template for auditing the agent's own actions during skill creation.
安全使用建议
This skill appears to implement a plausible meta-framework, but it also instructs the agent to modify and govern its own skills — a powerful capability that can change behavior persistently. Before installing or enabling: 1) Review the Python scripts (scripts/orchestrator.py, contract_validator.py, ecosystem_analyzer.py) for any network calls, arbitrary code execution, file write operations, subprocess uses, or reads of environment/credential files. 2) Run the skill in a tightly restricted sandbox (no write access to skill bundles or system-wide config) until you are confident of its behavior. 3) If you plan to allow it to run autonomously, ensure audit logging and human-in-the-loop gating for any actions that change skill code, registries, or escalate privileges. 4) Because the source/homepage is unknown, prefer a manual code review and consider limiting its permissions (filesystem, network, ability to modify other skills) or using a vetted alternative. If you want, I can summarize the orchestrator.py and other scripts line-by-line to highlight risky operations to inspect.
功能分析
Type: OpenClaw Skill
Name: intent-engineer
Version: 1.0.0
The bundle provides a complex meta-framework for agent orchestration and recursive self-improvement. It contains a significant security vulnerability in 'scripts/orchestrator.py', where the '_evaluate_condition' method uses 'eval()' to process condition strings, creating a risk of Remote Code Execution (RCE) if input is manipulated. Furthermore, the instructions in 'SKILL.md' and 'references/agent_decision_framework.md' encourage 'Intent Amplification' and autonomous framework modification, which are high-risk behaviors that could lead to the agent exceeding its intended scope or being exploited via prompt injection.
能力标签
能力评估
Purpose & Capability
Name, description, and provided resources (skill_registry.json, data contracts, governance docs, scripts/orchestrator.py) are coherent with a meta-framework for building and orchestrating skill ecosystems. The files present match the claimed purpose. One provenance note: the source/homepage is unknown which reduces trust but does not itself make the capability incoherent.
Instruction Scope
SKILL.md explicitly instructs the agent to use this framework to guide its own reasoning and to 'improve and extend the intent-engineering skill itself' (recursive self-modification). That is a broad, open-ended instruction granting the agent discretion to modify skills, register new skills, and orchestrate calls across the ecosystem. The instructions reference scripts/orchestrator.py and various registries — they do not declare explicit limits on what the agent may read or write, creating scope creep and risk of unintended changes to skill files or registries.
Install Mechanism
No install spec or external downloads are declared; the package is instruction-plus-bundled-files only. There are included Python scripts (scripts/*.py) but nothing in the metadata indicates remote code fetching or archive extraction. This lowers install-time supply-chain risk, but the presence of executable scripts means runtime execution could perform privileged actions if allowed.
Credentials
The skill declares no required environment variables or credentials, which is appropriate for a governance/orchestration framework. However, bundled Python scripts may still access environment variables, files, or network at runtime; SKILL.md does not enumerate or constrain what env/config the scripts may use. Because provenance is unknown, this is a cautionary signal rather than a confirmed mismatch.
Persistence & Privilege
always:false and normal autonomous invocation are set (no immediate red flag). But the skill's explicit goal of self-governance and recursive improvement implies it may modify registry files, templates, or skill files (which are included). The combination of executable scripts + instructions to update/extend the ecosystem suggest it could alter other skills or system artifacts if the agent has filesystem or config write permissions. That potential to persist changes across runs is a notable privilege to be aware of.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install intent-engineer - 安装完成后，直接呼叫该 Skill 的名称或使用
/intent-engineer触发 - 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
v1.0.0
- Initial release of the intent-engineering skill, introducing a meta-framework for orchestrating agent skills.
- Provides a structured operating system for agents to govern, align, and compose skills according to organizational goals.
- Implements an ecosystem architecture including: skill registry, data contracts, orchestration engine, shared intent framework, and agent decision framework.
- Establishes a 4-phase, ecosystem-aware workflow for building, registering, and orchestrating new skills.
- Offers new resources such as a shared intent guide, skill registry, data contract schemas, orchestration scripts, and audit templates.
元数据
常见问题
Intent-Engineering 是什么？
A meta-framework for designing, building, and orchestrating an ecosystem of strategically-aligned agent skills. This skill governs how the agent itself opera... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 79 次。
如何安装 Intent-Engineering？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install intent-engineer」即可一键安装，无需额外配置。
Intent-Engineering 是免费的吗？
是的，Intent-Engineering 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Intent-Engineering 支持哪些平台？
Intent-Engineering 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 Intent-Engineering？
由 Daniel Foo Jun Wei（@danielfoojunwei）开发并维护，当前版本 v1.0.0。
推荐 Skills