Thanks to visit codestin.com
Credit goes to github.com

Skip to content

[enhancement] Plan tier triggering criteria 太寬鬆 — 大範圍改動應 MUST-trigger Plan #129

Description

@kiki830621

Problem

Original feedback(user 2026-05-21,#44/#47 close 後反思):
「我覺得在 idd 裡面評斷要不要 plan 的標準太寬鬆了,如果大範圍的改動應該一定要啟動 plan」

現行 rules/sdd-integration.md Layer P 觸發 Plan tier 之條件為 disjunctive any-match(任一命中即觸發,但 default 仍是 Simple):

  • 2+ files have sequential dependency
  • 5+ ordered steps in strategy
  • Decision-heavy with 2+ valid approaches
  • Risk-sensitive boundary
  • Cross-file refactor without external contract change

User 提出問題:這些是「may trigger」criteria,沒有 hard MUST-trigger 防護;對於「大範圍改動」缺少強制升級到 Plan 的機制。

Type

enhancement(governance / Layer P 觸發判準調整)

Expected

引入 MUST-trigger layer 於 Layer P 之外或之上,讓「大範圍改動」一定走 Plan tier。可能的判準向度(plugin 維護者自行 calibrate):

  1. File-count threshold:預計修改 ≥ N 個檔案(N 可能 3、4、5,需 calibration)
  2. Structural / shared abstraction modifications:涉及共享資料結構、測量模型、helper interfaces、constants 等被多 caller 使用之 entity
  3. Cross-domain artifact dependency:改動 cascade 至多種 artifact 類型(e.g. R script + LaTeX + 表格 + 文檔)
  4. Family-scope changes:同類量表 / 同類組件之 family-wide 改動(本 session 例子:CR + PTSR + PCQ 三量表 LMI 修正皆屬同 family,但 feature: /idd-all chain-solve mode — 自動接續解 spawned issues + 單 review PR #44 只 plan 到 CR)

「Default Simple, escalate to Plan if Layer P fires」可能應改為「Default Plan if scope ≥ threshold, downgrade to Simple only if Layer 1 disqualifiers fire」之 inversion。

Actual — 來自 kiki830621/teaching-yun-2026 #44/#47 之 session evidence

#44 範例(2026-05-21):

  • Diagnosed as Plan tier(Layer P 觸發:多檔案順序依賴 + decision-heavy + risk-sensitive),走完整 plan-mode approval gate
  • 但 mid-implementation 出現 conceptual error(single-factor LMI 設計實質為 second-order factor disguise)— EnterPlanMode approval 沒抓到
  • User mid-thread correction「LMI 因素結構要和 §4.2.1 CFA 完全一樣」觸發 in-thread scope expansion
  • Plan scope 原本只談 CR,但實際擴及 CR + PTSR LMI(out-of-feature: /idd-all chain-solve mode — 自動接續解 spawned issues + 單 review PR #44-scope but consistency-driven)

#47 follow-up 必要性:

Pattern observation:#44 是「Plan 觸發了但 plan 本身 scope 不夠廣」的案例。User 提的「大範圍改動 MUST plan」可進一步要求 Plan tier scope 必須涵蓋 family-wide impact,不只當前 file 之最小修改。

Impact

  • rules/sdd-integration.md Layer P 評估邏輯
  • idd-diagnose Step 3.5 Complexity Assessment 之 routing decision
  • idd-plan Step 2 Implementation Plan 起草指引(可能需強制納入 family-wide scope check)
  • 不影響既有 Simple / Spectra tier 邊界

Priority

P2(process improvement;non-blocking 但 missing-Plan 的代價高 — per session evidence,#44 + #47 共需兩次 close 才完整覆蓋,若 #44 plan 一次到位可節省 ~30 分鐘 re-work)

Reference

Next

/idd-diagnose #N 探討具體 calibration:

  • File-count threshold 落在 N=?
  • 「Structural modification」如何 mechanically detectable(grep helper 修改、shared header import 等)?
  • Default flip(Simple → Plan)是否有 over-trigger 風險?Layer 1 disqualifiers 是否足夠 cover「真的 simple」的情境?

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions