135 lines
3.7 KiB
Markdown
135 lines
3.7 KiB
Markdown
---
|
|
artifact_type: repair-plan
|
|
name: ccpe-system-core-repair-plan
|
|
created: 2026-05-31
|
|
updated: 2026-05-31
|
|
status: draft
|
|
based_on: CCPE System
|
|
---
|
|
|
|
# CCPE System Core Repair Plan
|
|
|
|
## 1. Problem
|
|
|
|
The current CCPE System is strong at artifact governance: classifying, splitting, indexing, and maintaining Lite prompts, Agent Specs, Skills, Runtimes, and Model Cards.
|
|
|
|
It is weaker at preserving the original CCPE 2.0 prompt-construction kernel for mature single-agent expert prompts.
|
|
|
|
This creates two failure modes:
|
|
|
|
```text
|
|
1. Existing mature agents may be structurally upgraded but perform worse than their original CCPE 2.0 versions.
|
|
2. New deep expert agents may be over-agentified before their single-agent prompt core is strong enough.
|
|
```
|
|
|
|
## 2. Repair Principle
|
|
|
|
```text
|
|
Scenario first, artifact layers second.
|
|
Lite kernel first for single-agent expert use.
|
|
Agent / Skill / Runtime only when the scenario requires them.
|
|
```
|
|
|
|
## 3. Core Additions
|
|
|
|
### 3.1 Scenario Probe
|
|
|
|
Before creating, auditing, or refactoring an artifact, CCPE Forge must identify:
|
|
|
|
```text
|
|
- current or planned usage scenario
|
|
- target platform
|
|
- single-agent vs multi-agent use
|
|
- manual orchestration vs automation
|
|
- Web-like direct chat vs Codex callable behavior
|
|
- whether a Skill is needed for Codex invocation
|
|
- depth-oriented vs automation-oriented work
|
|
```
|
|
|
|
### 3.2 Mature Agent Minimal Expansion Rule
|
|
|
|
For mature, proven, single-agent expert prompts:
|
|
|
|
```text
|
|
default_output: Lite + optional Model Card
|
|
do_not_default_to: Agent + Skill + Runtime
|
|
```
|
|
|
|
Expand only when there is a real usage need:
|
|
|
|
```text
|
|
Skill: Codex callable method or reusable procedure
|
|
Agent: durable workflow role with collaboration contract
|
|
Runtime: multi-agent process, state, handoff, or automation
|
|
```
|
|
|
|
### 3.3 CCPE-Lite Construction Kernel
|
|
|
|
Lite is not merely a shortened Agent Spec.
|
|
|
|
For Web / GPT / Gemini / Claude style deployment, Lite should preserve the CCPE 2.0 working core:
|
|
|
|
```text
|
|
Core Layer
|
|
Execution Layer
|
|
Constraint Layer
|
|
Operation Layer
|
|
```
|
|
|
|
Use Outside-In construction:
|
|
|
|
```text
|
|
1. Alignment: purpose and usage scene
|
|
2. Scope: input, output, appendix / knowledge scope
|
|
3. Specification: output format and delivery standard
|
|
4. Core Construction: role, capability, constraints
|
|
5. Logic Design: workflow that produces the specified output
|
|
```
|
|
|
|
### 3.4 Regression Testing
|
|
|
|
When migrating a mature agent, validation must compare the new artifact against the old one:
|
|
|
|
```text
|
|
- Does the new Lite preserve the old agent's practical effect?
|
|
- Does it preserve the original voice and workflow?
|
|
- Does it improve defects without weakening production behavior?
|
|
- If tested, does it match or exceed old outputs in the intended scenario?
|
|
```
|
|
|
|
## 4. Files to Repair
|
|
|
|
```text
|
|
ccpe-protocol/ccpe-classification-rules.md
|
|
ccpe-protocol/ccpe-migration-policy.md
|
|
ccpe-protocol/ccpe-operating-modes.md
|
|
ccpe-protocol/ccpe-quality-rubric.md
|
|
.codex/skills/ccpe-forge/SKILL.md
|
|
.codex/skills/ccpe-forge/references/creator-mode.md
|
|
.codex/skills/ccpe-forge/references/refactor-mode.md
|
|
.codex/skills/ccpe-forge/templates/ccpe-lite.prompt.md
|
|
README.md
|
|
AGENTS.md
|
|
```
|
|
|
|
## 5. Non-Goals
|
|
|
|
```text
|
|
- Do not create the article review committee Runtime yet.
|
|
- Do not force Cognitive Imaging into Agent / Skill / Runtime in this repair pass.
|
|
- Do not rewrite the entire CCPE System.
|
|
- Do not change canonical model promotion in this pass unless separately requested.
|
|
```
|
|
|
|
## 6. Success Criteria
|
|
|
|
```text
|
|
- Scenario Probe appears in core workflow rules.
|
|
- Mature Agent Minimal Expansion Rule appears in classification and migration policy.
|
|
- CCPE-Lite template uses the four-layer CCPE 2.0 kernel.
|
|
- Quality Rubric evaluates scenario fit and Lite regression.
|
|
- Forge Creator and Refactor modes require scenario-based layer selection.
|
|
- Documentation warns against default four-layer expansion.
|
|
```
|
|
|