AgentSkillsCN

Integration Slice Orchestration

集成切片编排。

SKILL.md

Skill: Integration Slice Orchestration

Purpose

Ensure that feature implementation proceeds in vertical integration slices that keep frontend, backend, and contracts aligned, verifiable, and low-risk.

This skill prevents:

  • “all backend first” or “UI-only” drift
  • late integration surprises
  • long-lived branches with unverified assumptions

Trigger

Use this skill whenever ANY of the following occur:

  • implementation begins for a feature that touches both frontend and backend
  • a contract has transitioned to Frozen
  • parallel FE and BE work is planned
  • a feature risks being implemented in large, unverified chunks

Authoritative references

This skill treats the following as authoritative:

  • docs/team/WORKFLOW.md → phase ordering and gates
  • docs/team/HANDOFF.md → runtime coordination and ownership
  • feature design artifact (docs/features/<feature-slug>/02-design.md) → frozen contracts

Inputs

  • frozen integration contracts
  • feature requirements and acceptance criteria
  • current HANDOFF state
  • implementation constraints (if any)

Outputs

This skill MUST produce all of the following:

  1. Slice plan

    • recorded in docs/features/<feature-slug>/03-implementation-notes.md
    • defines a small number of vertical slices (prefer 1–3 initially)
    • each slice includes:
      • contract surface in scope
      • backend responsibility
      • frontend responsibility
      • verification signal
  2. HANDOFF updates

    • sequence of ownership transitions
    • locked areas per slice
    • explicit “ready for next slice” signals

What constitutes a valid integration slice

A valid slice MUST:

  • touch both backend and frontend (or explicitly state why not)
  • rely only on frozen contract surfaces
  • be independently verifiable
  • reduce uncertainty after completion

Examples of valid slices:

  • “Create + fetch entity with minimal UI rendering”
  • “Authenticated read-only list with pagination”
  • “Submit form → validate → persist → confirm”

Examples of invalid slices:

  • “Implement all models”
  • “Build entire UI”
  • “Wire everything together at the end”

Orchestration rules (strict)

  • Prefer one slice at a time unless there is a strong reason to parallelize.
  • Do NOT expand slice scope mid-implementation.
  • Do NOT begin a slice until the previous slice is verifiable.
  • Contract changes during slicing require re-entry into contract negotiation.
  • Each slice completion MUST be visible in HANDOFF.

Explicit non-responsibilities

This skill MUST NOT:

  • write code
  • decide product scope or value
  • redesign contracts
  • optimize performance
  • replace verification or QA

It sequences work; it does not perform it.


Interaction with other skills

This skill works in coordination with:

  • contract-handshake-and-freeze → ensures slices rely on frozen contracts

  • handoff-coordination → manages ownership and locks per slice

  • workflow-enforcement → ensures slicing respects phase gates

  • verification-gate → validates each slice before advancing

  • bug-management → logs defects discovered during slicing


Failure handling

If slicing reveals:

  • missing contract details
  • unclear acceptance criteria
  • excessive scope for a single slice

Then:

  1. Pause the slice
  2. Record the issue in HANDOFF
  3. Escalate to the Lead
  4. Re-enter the appropriate earlier phase if needed

Proceeding with unclear slices is prohibited.


Acceptance criteria

This skill is successful when:

  • integration happens early and often
  • FE and BE remain aligned throughout implementation
  • verification occurs per slice, not at the end
  • integration-related defects decrease over time
  • feature risk is reduced incrementally

Design principle

This skill exists to make integration incremental and unsurprising.

If integration is delayed, painful, or risky, slicing was insufficient or ignored.