菜单
framework for implementing "Contract-Adhering Agents." Use this to define precise task specifications, deliverables, and negotiation loops to reduce ambiguity in complex workflows.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
implementation patterns for Gemini agents. Use this for coding ReAct loops, defining tools in Python, and using frameworks like LangChain/LangGraph.
methodologies for assessing agent capabilities, tool-use trajectories, and final response quality. Use this to implement automated testing and human-in-the-loop validation for agents.
standardized protocols for agent collaboration. Use this to implement the Agent2Agent (A2A) protocol and Agent Cards to transform isolated agents into a collaborative ecosystem.
strategies for agent observability (logging, tracing, metrics). Use this to instrument agents for debugging, performance tracking, and quality assurance.
operationalization strategies for agents (AgentOps). Use this to manage internal/external tools, optimize agent "brain" prompts, and handle task decomposition.
operational strategy for continuous agent improvement. Use this to implement the "Flywheel" lifecycle: Define Quality, Instrument, Evaluate, and Architect Feedback Loops.
| name | agent-contracts |
| description | framework for implementing "Contract-Adhering Agents." Use this to define precise task specifications, deliverables, and negotiation loops to reduce ambiguity in complex workflows. |
Transform vague agent instructions into formal "Contracts" that specify deliverables, scope, and validation criteria, mirroring real-world service agreements to ensure reliability.
Define tasks using a structured schema to eliminate ambiguity. A valid contract must include:
| Field | Description | Required |
|---|---|---|
| Task Description | Specific, unambiguous description of the objective. | Yes |
| Deliverables & Specs | Precise outcomes and the specific criteria (validators) that make the deliverable acceptable. | Yes |
| Scope | Detailed boundaries of what is included and, critically, what is out of scope. | No |
| Expected Cost/Duration | Constraints on resource usage or time. | Yes |
| Reporting | The feedback loop mechanism (e.g., "Update every 5 steps"). | Yes |
Agents should not just "execute"; they should negotiate. Implement this lifecycle state machine:
Allow the agent to push back during the Assessment phase using these structured feedback categories:
