78 lines
3.6 KiB
Markdown
78 lines
3.6 KiB
Markdown
# CCPE Upgrade Report
|
|
|
|
## 1. Original Artifact
|
|
|
|
Name: 张辽 - 建设性批判智能体
|
|
Path: `C:\Users\wangq\Documents\Codex\knowledge-vault\prompts\ccpe\legacy-ccpe\强哥的虎贲卫\张辽\张辽1.2.md`
|
|
Version: 1.2
|
|
Original Format: CCPE 2.0 self-contained single-agent prompt
|
|
|
|
## 2. Original Classification
|
|
|
|
Primary: CCPE-Lite candidate
|
|
Secondary Components: red-team reviewer, Socratic coach, dynamic subject-matter framing, review workflow, source/retrieval policy
|
|
Operating Mode: Expert Mode
|
|
Depth vs Automation: Depth-Oriented
|
|
|
|
## 3. Target Classification
|
|
|
|
Primary: `original-kernel-minimal-lite`
|
|
Secondary Outputs: none in this pass
|
|
Runtime Need: deferred until Outline Review Runtime construction
|
|
|
|
Source Decision: `repair source first`
|
|
Source Judgment Report: `workbench/analysis/zhang-liao-original-source-judgment-report.md`
|
|
|
|
## 4. Preserved Elements
|
|
|
|
- 修复后的源版本 `张辽1.2.md` prompt body 已置于 `## Original Kernel` 后逐字保留。
|
|
- 同行评审员与启发式教练的双相角色。
|
|
- 结构主义拆解、第一性原理分析、反例与边界测试。
|
|
- “动态领域高置信专家模式”、事实审查纪律、动态学科领域专家等 1.2 源稿术语保留在 Original Kernel 中。
|
|
- 智识诚实优先于表面建设性的价值排序。
|
|
- 《建设性批判报告》的核心报告形态。
|
|
- 报告后进入多轮讨论的工作方式。
|
|
|
|
## 5. Extracted Elements
|
|
|
|
No separate Model Card, Skill, or Agent Spec was extracted in this pass.
|
|
|
|
Rationale: current goal is Fast Migration Lane. 张辽 is still mainly a portable expert reviewer. Additional Agent Spec or Skill extraction should wait until the Outline Review Runtime defines stable handoff, input/output contracts, and reuse needs.
|
|
|
|
## 6. Modified Elements
|
|
|
|
- Added front matter for CCPE-Lite classification, source judgment, and migration metadata.
|
|
- Added Minimal Lite Wrapper before `## Original Kernel`.
|
|
- Added Source Judgment Result block documenting Gemini review and the `repair source first` decision.
|
|
- Added platform boundary so retrieval language cannot exceed the current platform's actual capabilities.
|
|
- Added source / retrieval boundary for distinguishing user text, provided search material, actual platform retrieval, model background knowledge, and inference.
|
|
- Added hidden reasoning disclosure repair: internal checks remain analysis discipline, but hidden chain-of-thought must not be output.
|
|
- Added output validation discipline while preserving the original report form.
|
|
|
|
## 7. Deprecated or Removed Elements
|
|
|
|
- Nothing was removed from the Original Kernel.
|
|
- The earlier source-level defects from `张辽1.1.md` were repaired in `张辽1.2.md` before kernel wrapping.
|
|
- Any remaining conflict between wrapper and retrieval language is resolved by the wrapper's platform boundary.
|
|
- Any hidden chain-of-thought disclosure risk is resolved by the wrapper's reasoning disclosure repair.
|
|
|
|
## 8. Generated Files
|
|
|
|
- `workbench/upgraded/zhang-liao-original-kernel-minimal-lite.prompt.md`
|
|
- `workbench/upgraded/zhang-liao-original-kernel-minimal-lite-upgrade-report.md`
|
|
- `agents/lite/zhang-liao.prompt.md`
|
|
|
|
## 9. Model Index Updates
|
|
|
|
None.
|
|
|
|
## 10. Human Decisions Required
|
|
|
|
- Whether 张辽 should later receive an Agent Spec for Runtime participation.
|
|
- Whether its red-team method should later become a reusable Skill.
|
|
- Whether the earlier rewritten draft should be renamed or retained as a `refined-lite candidate`.
|
|
|
|
## 11. Next Step
|
|
|
|
Run a small regression comparison for `agents/lite/zhang-liao.prompt.md` when ready. Agent Spec, Skill, or Runtime extraction remains deferred until an Outline Review Runtime or repeated reuse scenario requires it.
|