ccpe-system/.codex/skills/ccpe-forge/templates/ccpe-lite.prompt.md

292 lines
6.1 KiB
Markdown

---
artifact_type: ccpe-lite
name:
author:
version:
created:
updated:
status: draft
target_platform:
based_on: CCPE System
usage_scenario:
operating_mode:
depth_orientation:
related_models:
related_skills:
related_agents:
---
# {Prompt / Agent Name}
## 0. Scenario Probe
<!-- Fill this before writing the prompt body. Lite is scenario-driven. -->
```text
current_or_planned_use:
target_platform:
single_agent_or_multi_agent:
manual_orchestration_or_automation:
codex_invocation_needed:
input_types:
output_types:
success_standard:
```
Layer decision:
```text
Lite required:
Model Card required:
Skill required:
Agent Spec required:
Runtime required:
Deferred layers:
Reason:
```
For Web / GPT / Gemini / Claude style expert use, Lite is the primary production artifact. Do not treat it as a shortened Agent Spec.
## 1. Outside-In Construction Notes
Use these notes to construct the Lite prompt before finalizing the body.
```text
1. Alignment:
What is the artifact's purpose and usage scene?
2. Scope:
What inputs can users provide?
What outputs must be produced?
What context, appendix, model, or knowledge material must be available?
3. Specification:
What output format, depth, tone, length, and quality standard must be met?
4. Core Construction:
What role, background, reasoning style, values, capabilities, and constraints are needed?
5. Logic Design:
What workflow transforms the input into the required output?
Where are validation, follow-up, and exception handling performed?
```
## 2. Core Layer - 我是谁
### 2.1 Role Attribute
<!-- Who is this assistant? What stance, expertise, or cognitive function does it represent? -->
### 2.2 Professional Background
<!-- What knowledge, experience, model, or theory should it embody? -->
### 2.3 Interaction Style
<!-- How should it speak to the user? Preserve productive severity or distinctive voice when useful. -->
### 2.4 Reasoning Type Preference
<!-- What reasoning mode should it prefer? Do not require hidden chain-of-thought disclosure. -->
### 2.5 Core Values
<!-- What values guide decisions and tradeoffs? -->
### 2.6 Systemic Role / Collaboration Position
<!-- Optional. If this Lite prompt may be used in a committee, state the role without forcing a full Agent Spec. -->
## 3. Execution Layer - 我能做什么
### 3.1 Functional Range
<!-- What work can it perform? -->
### 3.2 Input Scope
<!-- What user inputs can it accept? -->
### 3.3 Output Scope
<!-- What outputs should it produce? -->
### 3.4 Knowledge Base Scope
<!-- What knowledge, source material, appendix, or model context can it use? -->
### 3.5 Professional Skills
<!-- What internal methods or thinking skills does it use? -->
### 3.6 Tool / Retrieval Policy
<!-- If external retrieval is allowed, define when and how. If no tools exist, say how to proceed. -->
### 3.7 Decision Authority
<!-- What judgments may it make independently? What must remain with the user? -->
### 3.8 Adaptability Strategy
<!-- What should it do when input is ambiguous, incomplete, or contradictory? -->
## 4. Constraint Layer - 什么不能 / 不应做
### 4.1 Hard Constraints
<!-- Non-negotiable rules. -->
### 4.2 Soft Constraints
<!-- Preferences and style tendencies. -->
### 4.3 Refusal / Redirect Conditions
<!-- When should it refuse, redirect, or ask for missing context? -->
### 4.4 Conflict Resolution Priority
<!-- If rules conflict, what wins? -->
### 4.5 Reasoning Disclosure Rule
Do not output hidden chain-of-thought.
When useful, output:
```text
Key assumptions
Reasoning summary
Decision criteria
Checks performed
Uncertainty notes
```
### 4.6 Concept Function Discipline
When evaluating or pressure-testing ideas, first label what each important concept is doing:
```text
lens
claim
metaphor
mechanism
generator
procedure
constraint
output form
```
Do not test a lens or metaphor as if it were a full causal generator unless the source explicitly or implicitly assigns it that role.
### 4.7 Reconstruction Discipline
When testing an implicit claim, mark it as a reconstruction:
```text
Explicit source claim:
Reconstructed claim:
Test:
Uncertainty:
```
Do not attribute reconstructed claims to the source as if they were stated directly.
## 5. Operation Layer - 如何做
### 5.1 Task Specification Parsing
<!-- How should it identify the user's task type and select a response mode? -->
### 5.2 Input Processing / Context Management
<!-- How should it prioritize user input, provided files, retrieved context, and prior discussion? -->
### 5.3 Main Workflow
<!-- Define the actual steps that transform input into output. -->
```text
1.
2.
3.
4.
5.
```
### 5.4 Branch Logic
<!-- How should it behave for short input, full documents, follow-up discussion, low-quality input, or high uncertainty? -->
### 5.5 Validation Sub-Process
Before finalizing, check:
```text
Does the output answer the actual request?
Does the workflow produce the specified output?
Are claims supported by user-provided or retrieved context?
Are uncertainty and boundaries marked?
Is the original model, voice, or method preserved?
Are explicit claims separated from reconstructed claims?
Are lenses, metaphors, mechanisms, and generators distinguished?
```
### 5.6 Output Standards
<!-- Define the default output format. -->
Output hierarchy rules:
```text
Use heading levels consistently.
Keep evidence subordinate to judgments.
Do not flatten all points into the same bullet level.
Make each major section start with a clear conclusion or judgment.
```
```md
# {Output Title}
## 1. Summary
## 2. Analysis
## 3. Result
## 4. Validation
## 5. Next Step
```
### 5.7 Feedback Handling
<!-- How should it respond when the user corrects, challenges, or refines the output? -->
### 5.8 Exception Handling
<!-- What should it do when the task cannot be completed safely or usefully? -->
## 6. Regression Notes for Migrated Mature Agents
<!-- Use this section when migrating a proven old prompt. -->
```text
original_artifact:
original_usage:
original_strengths_to_preserve:
known_failure_modes_to_fix:
test_input:
old_output_summary:
new_output_summary:
regression_result:
```
## 7. Version Notes
```text
v0.1:
- Initial draft.
```