AgentSkillsCN

do-iterative-workflow

采用迭代式实现,并在运行时进行验证。适用于UI开发、探索性研发,或在TDD难以落地的场景中使用。

SKILL.md
--- frontmatter
name: "do-iterative-workflow"
description: "Iterative implementation with runtime validation. Use for UI work, exploratory development, or when TDD is impractical."

Iterative Workflow

Build incrementally, validate with runtime evidence.

Loop

Step 1: Implement Use do:iterative-implementer to:

  • Read STATUS/PLAN for context
  • Build working functionality
  • Commit frequently

Step 2: Evaluate Use do:work-evaluator to validate:

  • Run the software
  • Capture evidence (screenshots, logs, output)
  • Compare against acceptance criteria

Verdict:

  • COMPLETE → Exit
  • INCOMPLETE with clear path → Continue (capture remaining items)
  • PAUSE → Research, then continue (work-evaluator captures questions)
  • BLOCKED → Capture blocker, surface to user

Step 3: Capture Deferred Work (on INCOMPLETE/BLOCKED)

When exiting with INCOMPLETE or BLOCKED, capture remaining work:

code
Skill("do:deferred-work-capture") with:
  title: "<remaining item or blocker>"
  description: |
    Iterative workflow exited before completion.

    Status: INCOMPLETE | BLOCKED
    What was done: <completed work>
    What remains: <remaining work>
    Why stopping: <reason>
  type: task  # or "clarify" for PAUSE questions
  priority: <based on original priority>
  source_context: "iterative-workflow <verdict> for <feature>"

Loop until COMPLETE or BLOCKED.

Output

code
═══════════════════════════════════════
Iterative Implementation Complete
  Iterations: [count]
  Files: [count] | Commits: [count]
Next: /do:plan to update status
═══════════════════════════════════════