AgentSkillsCN

collaboration-flow

文档更新的人机协作流程。关键词:协作、工作流程、更新。

SKILL.md
--- frontmatter
name: collaboration-flow
description: "Human-AI collaboration flow for document updates. Keywords: collaboration, workflow, updates."

AI-Human Collaboration Flow

Standard Collaboration Pattern

The canonical workflow is: AI Draft -> Human Review -> AI Revise

mermaid
flowchart LR
    A[AI Drafts] --> B[Human Reviews]
    B -->|Approved| C[AI Finalizes]
    B -->|Feedback| D[AI Revises]
    D --> B
    C --> E[Outcome Recorded]

Decision Recording

Decision TypeRecord LocationPromotion Target
Task-specific decisionsworkdocs/ (plan.md, context.md)Archive when complete
Stable knowledgeworkdocs/ initiallyPromote into the relevant skill package under /.system/skills/ssot/**
Repeated patternsworkdocs/ observationsAbilities or workflows
Policy decisionsReview commentsUpdate AGENTS.md if applicable

Review Feedback Format

Humans provide review feedback using inline comments or structured blocks:

markdown
<!-- REVIEW: [reviewer_name] [date]
Status: approved | needs_revision | blocked
Feedback: Specific feedback text here
-->

AI must:

  1. Parse and address all needs_revision feedback.
  2. Not proceed past blocked status without human resolution.
  3. Update workdocs with review outcomes.

Human Approval Gates

Certain actions require explicit human approval:

Action TypeApproval RequiredRecord Method
Production deploymentYesworkdocs/tasks.md checkpoint
Security-sensitive changesYesworkdocs/tasks.md checkpoint
Cross-module breaking changesYesIntegration scenario workdocs
Credential/secret accessYesDevOps scenario workdocs

AI must stop and request approval when:

  • A hook returns requires_human.
  • An action is in the "Forbidden without approval" list in applicable AGENTS.md.
  • Uncertainty exists about safety implications.

Related documents