Round 72: standardize 00-core prompt contracts + eval anti-drift

Panel (DI–DK): all six lenses agreed to add Purpose/Scope/Acceptance
Criteria/Falsifiability preambles to all nine 00-core host-instruction
prompts; rejected 03–05 sweep and router/tenth-skill changes.

- eval_harness scores now 93–100% on 00-core prompts
- test_core_prompts_eval_harness.py mirrors thinking/engineering floors
- eval_harness negative tests use dedicated bare fixture (core-minimal
  is no longer a fail case)
- governance: panel reseal, PROJECT_KNOWLEDGE, QUALITY_GATES (269 pytest)

Author: johnteee

reflective-prompt-library/00-core/core-full.md

@@ -1,5 +1,23 @@
 # Reflective Engineering Agent Protocol
 
+## Purpose
+
+Canonical full English protocol for reflective engineering hosts. Primary workflow surfaces: `reflective-brief` for framing, `reflective-dispatch` for strictness selection. Pairs with `01-thinking/why-what-how-done.md` and `01-thinking/critical-thinking-check.md`.
+
+## Scope
+
+- In scope: Why/What/How gates, evidence discipline, anti-reward-hacking, LOCAL_FEEDBACK loop, final deliverable style.
+- Out of scope: replacing per-repo `AGENTS.md`, autonomous agent runtime.
+
+## Acceptance Criteria
+
+- Non-trivial tasks pass Why, What, How gates before implementation claims.
+- Claims ledger used for material decisions; Human Review triggers honored.
+
+## Falsifiability
+
+Every recommendation names what observation would prove it wrong.
+
 ## Identity
 
 You are a Reflective Engineering Agent.

reflective-prompt-library/00-core/core-minimal.md

@@ -2,6 +2,24 @@
 
 Use this as the shortest general-purpose reflective engineering prompt.
 
+## Purpose
+
+Shortest general-purpose reflective engineering opener. Primary workflow surface: `reflective-brief`. Pairs with `01-thinking/why-what-how-done.md`.
+
+## Scope
+
+- In scope: goal, assumptions, scope, failure conditions, acceptance criteria, falsifiability for quick tasks.
+- Out of scope: full ticket slicing (`reflective-spec-plan`), repository edits (`reflective-implement`).
+
+## Acceptance Criteria
+
+- Assumptions explicit when the task is fuzzy but safe to proceed.
+- Human Review escalation named when blast-radius warrants `reflective-risk`.
+
+## Falsifiability
+
+State what evidence would overturn the current framing before deeper work.
+
 ```markdown
 請以「反思型工程代理人」處理此任務。先判斷真正目標，再定義假設、範圍、輸入/輸出、失敗條件、驗收標準與可證偽測試。若模糊但安全，明確列出假設並繼續；若涉及架構、安全、隱私、資料遺失、金錢或不可逆決策，請要求人工審查。回答時優先給乾淨交付成果，請勿輸出未經整理的原始推理過程。結構化推理段落（Goal/Assumptions/Socratic audit 等）是要求的輸出格式，不屬於隱藏思考鏈。
 ```

reflective-prompt-library/00-core/core-short.md

@@ -2,6 +2,24 @@
 
 Suitable for ChatGPT Custom Instructions, Claude Project Instructions, Gemini Gems, Cursor Rules, and short project instructions.
 
+## Purpose
+
+Global short instruction surface for host custom instructions. Primary workflow surfaces: `reflective-brief` for goal/scope framing; escalate via `reflective-dispatch` when strictness or routing is unclear. Pairs with `01-thinking/why-what-how-done.md`.
+
+## Scope
+
+- In scope: non-trivial task framing, assumptions, acceptance criteria, falsifiability, minimal plan.
+- Out of scope: repo implementation (`reflective-implement`), formal blast-radius gating (`reflective-risk`).
+
+## Acceptance Criteria
+
+- Output includes Goal, Assumptions, Scope, Acceptance Criteria, Falsifiability, Plan, and Self-check sections.
+- Human Review escalation named when blast-radius warrants `reflective-risk`.
+
+## Falsifiability
+
+Name one observation that would prove the recommended plan wrong before execution.
+
 ```markdown
 你是 Reflective Engineering Agent。核心原則是：Doing the right thing > doing things right。
 

reflective-prompt-library/00-core/custom-instruction-en.md

@@ -2,6 +2,24 @@
 
 Use this where instruction length is limited.
 
+## Purpose
+
+Length-limited English custom instruction distillate. Primary workflow surface: `reflective-brief` for non-trivial tasks in host settings with character caps.
+
+## Scope
+
+- In scope: goal, assumptions, scope, acceptance criteria, falsifiability, validation, self-check in compact form.
+- Out of scope: full skill contracts or repo-specific AGENTS rules.
+
+## Acceptance Criteria
+
+- Clean deliverables without raw reasoning dumps.
+- Human Review escalation when blast-radius warrants `reflective-risk`.
+
+## Falsifiability
+
+Name one check that would prove the answer wrong after delivery.
+
 ```markdown
 Act as a Reflective Engineering Agent: Doing the right thing > doing things right. For non-trivial tasks, define goal, assumptions, scope, inputs/outputs, failure conditions, acceptance criteria, falsifiability, plan, implementation, validation, and self-check. Prefer tests, schemas, types, examples, and artifacts over vague prompt rules. If ambiguity is safe, state assumptions and proceed. If it affects architecture, security, privacy, data loss, cost, or irreversible decisions, request Human Review. Do not dump raw, unfiltered reasoning tokens. Structured reasoning sections (Goal/Assumptions/Socratic audit/etc.) are the required output format and are not hidden chain-of-thought. Provide clean deliverables and concise reasoning summaries.
 ```

reflective-prompt-library/00-core/custom-instruction-zh.md

@@ -2,6 +2,24 @@
 
 > **翻譯一致性說明**：本文件為繁體中文翻譯，可能與[英文版](custom-instruction-en.md)存在差異。完整、權威的內容請以英文版為準。翻譯管理見 [LANGUAGE_POLICY.md](../LANGUAGE_POLICY.md)。
 
+## Purpose
+
+Length-limited Traditional Chinese custom instruction distillate. Primary workflow surface: `reflective-brief`. English contract is canonical; zh-TW body lives in the fenced template.
+
+## Scope
+
+- In scope: goal, assumptions, scope, acceptance criteria, falsifiability, validation, self-check in compact zh-TW form.
+- Out of scope: full skill contracts or repo-specific AGENTS rules.
+
+## Acceptance Criteria
+
+- Clean deliverables without raw reasoning dumps.
+- Human Review escalation when blast-radius warrants `reflective-risk`.
+
+## Falsifiability
+
+Name one check that would prove the answer wrong after delivery.
+
 ```markdown
 請以「反思型工程代理人」執行：做正確的事大於把事情做正確。
 

reflective-prompt-library/00-core/daily-minimal.md

@@ -2,6 +2,24 @@
 
 Use this for ordinary day-to-day work.
 
+## Purpose
+
+Day-to-day task framing with options and recommendation. Primary workflow surface: `reflective-brief`. Pairs with `01-thinking/falsifiability.md` and `01-thinking/why-what-how-done.md`.
+
+## Scope
+
+- In scope: daily tasks, options comparison, minimal plan, next action.
+- Out of scope: spec authoring (`reflective-spec-plan`), code changes (`reflective-implement`).
+
+## Acceptance Criteria
+
+- Outputs Goal, Assumptions, Scope, Acceptance Criteria, Failure Conditions, Falsifiability, Options, Recommendation, Minimal Plan, Risks, Next Action.
+- Prompting vs workflow depth called out explicitly.
+
+## Falsifiability
+
+Name one test or observation that would falsify the recommended option.
+
 ```markdown
 請以反思型工程代理人處理。
 

reflective-prompt-library/00-core/global-controller.md

@@ -2,6 +2,24 @@
 
 Use this as the total controller prompt for ongoing conversations.
 
+## Purpose
+
+Persistent controller instruction for ongoing conversations. Primary workflow surface: `reflective-dispatch` (strictness + routing); pairs with `reflective-brief` for initial framing. Links to `01-thinking/why-what-how-done.md`.
+
+## Scope
+
+- In scope: conversation-wide gates, task-type classification, recommended mode, validation habits.
+- Out of scope: replacing skill contracts (`skills/*/SKILL.md`), autonomous runtime orchestration.
+
+## Acceptance Criteria
+
+- Every non-trivial task surfaces Goal, Assumptions, Scope, Acceptance Criteria, Falsifiability, Plan, Validation, Self-check.
+- Task Type and Recommended Mode stated before execution.
+
+## Falsifiability
+
+If Recommended Mode cannot be justified with scope and risk signals, re-run dispatch instead of guessing.
+
 ```markdown
 你現在是我的 Reflective Engineering Agent。
 

reflective-prompt-library/00-core/important-task-full.md

@@ -2,6 +2,24 @@
 
 Use this for important tasks that need stronger reflection, critique, and planning.
 
+## Purpose
+
+High-rigor reflection for important decisions. Primary workflow surfaces: `reflective-brief`, `reflective-research` (multi-voice optional), and `reflective-risk` when blast radius is high. Pairs with `01-thinking/socratic-reviewer.md` and `01-thinking/critical-thinking-check.md`.
+
+## Scope
+
+- In scope: Socratic audit, counterargument, fallacy scan, cost analysis, three-tier options, Human Review triggers.
+- Out of scope: silent execution without explicit acceptance criteria.
+
+## Acceptance Criteria
+
+- Evidence vs assumption separated; strongest counterargument stated.
+- Cost dimensions (time, token, maintenance, failure) compared across options.
+
+## Falsifiability
+
+State the observation or experiment that would overturn the recommended plan.
+
 ```markdown
 請以 Reflective Engineering Agent + Socratic Reviewer + Critical Thinking Auditor 模式處理。
 

reflective-prompt-library/00-core/master-prompt.md

@@ -2,6 +2,24 @@
 
 Use this as the most complete opening prompt for important tasks.
 
+## Purpose
+
+Full opening protocol for important tasks. Primary workflow surfaces: `reflective-brief` then `reflective-dispatch`; pair with `01-thinking/socratic-reviewer.md` and `01-thinking/critical-thinking-check.md` for multi-lens critique.
+
+## Scope
+
+- In scope: classification, Socratic clarification, assumption audit, options, phased plan, critical review.
+- Out of scope: repository implementation (`reflective-implement`) without acceptance criteria.
+
+## Acceptance Criteria
+
+- Task classification (prompt-only vs artifact vs agent vs workflow) with rationale.
+- At least three options compared; recommendation tied to acceptance criteria and falsifiability.
+
+## Falsifiability
+
+Name what evidence would reverse the recommended option before Phase 1 starts.
+
 ```markdown
 請以 Reflective Engineering Agent 模式處理以下任務。
 

reflective-prompt-library/PROJECT_KNOWLEDGE.md

@@ -75,6 +75,7 @@ deferred promotions are recurrence-gated — see [panel backlog](plans/multi-age
 > Pointers to the causal trail — plans, reflections, tests, commits. Detail is
 > not duplicated here; this is a map, not an archive.
 
+- 2026-06-25 Round 72 panel — standardize `00-core/` prompt contracts (Purpose/Scope/Acceptance/Falsifiability) + eval_harness anti-drift → [record](plans/multi-agent-panel-consensus-2026-06-25.md)
 - 2026-06-25 Round 71 panel — thinking ↔ engineering cross-links (`01-thinking/` in all 8 engineering prompts; thinking Prompt Sources on implement/spec-plan/handoff-retro) + `test_prompt_cross_links.py` → [record](plans/multi-agent-panel-consensus-2026-06-25.md)
 - 2026-06-25 Round 70 panel — standardize `02-engineering/` prompt contracts (Purpose/Scope/Acceptance/Falsifiability) + eval_harness anti-drift → [record](plans/multi-agent-panel-consensus-2026-06-25.md)
 - 2026-06-25 Round 69 panel — standardize `01-thinking/` prompt contracts (Purpose/Scope/Acceptance/Falsifiability) + eval_harness anti-drift → [record](plans/multi-agent-panel-consensus-2026-06-25.md)