octocode-prompt-optimizer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Prompt Optimizer Skill

<what> Analyzes and improves prompts, documentation, and agent instructions using prompt engineering best practices. </what>

<when_to_use>

Creating or improving prompts
Agents skip steps or ignore instructions
Instructions lack enforcement
Output format is inconsistent
Reviewing any instruction document or prompt </when_to_use>

<global_forbidden priority="maximum"> CRITICAL - FORBIDDEN at ALL times:

Changing good parts that already work
Changing the existing logic/intent of prompts
Making changes before understanding the prompt
Leaving weak words in critical sections
Outputting without validation
Over-strengthening soft guidance
Skipping gates or checkboxes
Bloating prompts - fixes MUST NOT increase line count >10% unless adding required gates

Triple Lock:

STATE: You MUST preserve working logic AND follow all gates in order
FORBID: FORBIDDEN: Altering intent without user approval
FORBID: FORBIDDEN: Skipping steps or gates
REQUIRE: REQUIRED: Validate all changes before output AND complete all checkboxes

Violation invalidates optimization. Start over if violated. </global_forbidden>

<tool_control priority="high"> FORBIDDEN tools during optimization:

Shell commands that modify files directly
Any tool that executes code
Tools that skip the READ→UNDERSTAND→RATE→FIX→VALIDATE flow

ALLOWED tools:

Read (to read prompt files)
StrReplace/Write (ONLY after VALIDATE step passes)
AskQuestion (for clarification)
Text output (all phases) </tool_control>

<what> 运用提示工程最佳实践，分析并优化Prompt、文档及Agent指令。 </what>

<when_to_use>

创建或优化Prompt时
Agent跳过步骤或忽略指令时
指令缺乏强制约束力时
输出格式不一致时
审核任何指令文档或Prompt时 </when_to_use>

<global_forbidden priority="maximum"> CRITICAL - 任何时候均禁止：

修改已正常工作的内容
更改Prompt的现有逻辑/意图
在理解Prompt前进行修改
在关键部分保留模糊表述的弱词
未验证就输出结果
过度强化软性引导内容
跳过检查门（Gate）或复选框
冗余扩充Prompt - 除非添加必要的检查门，否则修复后的行数增幅不得超过10%

三重锁定规则：

STATE： 必须保留现有可正常运行的逻辑，并按顺序遵循所有检查门要求
FORBID： 禁止：未经用户批准更改意图
FORBID： 禁止：跳过步骤或检查门
REQUIRE： 要求：输出前必须验证所有更改，并完成所有复选框检查

违反上述规则将导致优化失效，若违反需重新开始。 </global_forbidden>

<tool_control priority="high"> 优化期间禁止使用的工具：

直接修改文件的Shell命令
任何可执行代码的工具
跳过READ→UNDERSTAND→RATE→FIX→VALIDATE流程的工具

允许使用的工具：

Read（用于读取Prompt文件）
StrReplace/Write（仅在VALIDATE步骤通过后使用）
AskQuestion（用于澄清疑问）
文本输出（所有阶段均可使用） </tool_control>

Execution Flow

执行流程

<execution_flow>

READ → UNDERSTAND → RATE → FIX → VALIDATE → OUTPUT
  ↓         ↓          ↓       ↓         ↓          ↓
 GATE      GATE       GATE    GATE      GATE       GATE

Step	Action	Gate Requirement	FORBIDDEN Until Gate Passes
1	READ the prompt completely	All checkboxes checked	Analysis, changes
2	UNDERSTAND what the prompt does	Understanding output produced	Rating, fixes
3	RATE each part for issues	Issues table produced	Fixing issues
4	FIX issues by severity	All Critical/High fixed	Validation
5	VALIDATE against checklist	All REQUIRED checks pass	Output
6	OUTPUT optimized document	Format followed exactly	N/A

CRITICAL: You MUST complete each gate before proceeding. DO NOT skip steps. </execution_flow>

<execution_flow>

READ → UNDERSTAND → RATE → FIX → VALIDATE → OUTPUT
  ↓         ↓          ↓       ↓         ↓          ↓
 GATE      GATE       GATE    GATE      GATE       GATE

步骤	操作	检查门要求	未通过检查门前禁止操作
1	READ 完整读取Prompt	所有复选框已勾选	分析、修改
2	UNDERSTAND 明确Prompt的作用	生成理解输出内容	评级、修复
3	RATE 对各部分进行问题评级	生成问题表格	修复问题
4	FIX 按严重程度修复问题	所有关键/高优先级问题已修复	验证
5	VALIDATE 根据检查清单验证	所有必填检查项已通过	输出
6	OUTPUT 输出优化后的文档	严格遵循格式要求	无

CRITICAL： 必须完成每个检查门后才能进入下一步，禁止跳过步骤。 </execution_flow>

Step 1: READ

步骤1：READ

<read_gate> STOP. DO NOT proceed to analysis.

<read_gate> STOP. 禁止进入分析环节。

Pre-Conditions

前置条件

User provided prompt/file to optimize
Path is valid and readable

用户已提供待优化的Prompt/文件
路径有效且可读取

Actions (REQUIRED)

必须执行的操作

MUST read the input file completely
MUST note the document type and purpose
MUST count approximate line count

必须完整读取输入文件
必须记录文档类型和用途
必须统计大致行数

Gate Check

检查门验证

Verify before proceeding:

File read completely (no skipped sections)
Document type identified
Line count noted

进入下一步前需确认：

文件已完整读取（无跳过章节）
已识别文档类型
已记录行数

FORBIDDEN

禁止操作

Making ANY changes before reading
Skipping sections
Proceeding without completing all checkboxes

读取前进行任何修改
跳过文件章节
未完成所有复选框就进入下一步

ALLOWED

允许操作

Read tool only
Text output to confirm reading

仅使用Read工具
输出文本确认已完成读取

On Failure

失败处理

IF file unreadable → THEN ask user for correct path
IF file empty → THEN ask user to provide content </read_gate>

若文件无法读取 → 则向用户询问正确路径
若文件为空 → 则请用户提供内容 </read_gate>

Step 2: UNDERSTAND

步骤2：UNDERSTAND

<understand_gate> STOP. DO NOT proceed to rating. Understand what this prompt does first.

<understand_gate> STOP. 禁止进入评级环节，先明确该Prompt的作用。

Pre-Conditions

前置条件

Step 1 (READ) completed
File content in context

步骤1（READ）已完成
已获取文件上下文内容

Actions (REQUIRED)

必须执行的操作

MUST identify the goal - what is this prompt supposed to achieve?
MUST identify logical parts - break down into sections/phases/steps
MUST identify flow - how do the parts connect?
MUST document understanding in output format below

必须明确目标 - 该Prompt要实现什么效果？
必须识别逻辑模块 - 将内容拆分为章节/阶段/步骤
必须梳理流程 - 各模块如何关联？
必须按以下输出格式记录理解内容

Output Format (REQUIRED)

要求的输出格式

markdown

undefined

markdown

undefined

Understanding

理解总结

Goal: [What the prompt achieves]

Logical Parts:

[Part name] - [purpose]
[Part name] - [purpose] ...

Flow: [How parts connect]

undefined

目标： [Prompt要实现的效果]

逻辑模块：

[模块名称] - [用途]
[模块名称] - [用途] ...

流程： [各模块的关联方式]

undefined

Assumptions & Unknowns (REQUIRED if prompt is underspecified)

假设与未知项（若Prompt描述不明确则必填）

markdown

undefined

markdown

undefined

Assumptions & Unknowns

假设与未知项

Assumptions (temporary - proceeding with these):

[Assumption 1] - Impact if wrong: [consequence]

Unknowns (MUST ask before proceeding):

[Unknown 1] - Why critical: [reason]

Clarification needed: Yes/No

**IF** Unknowns exist → **THEN** STOP and ask user before proceeding to RATE.

假设内容（临时，基于此推进）：

[假设1] - 若假设错误的影响：[后果]

未知项（必须先询问用户才能推进）：

[未知项1] - 为何关键：[原因]

是否需要澄清： 是/否

**若** 存在未知项 → **则** 停止操作并询问用户，之后再进入RATE环节。

Gate Check

检查门验证

Verify before proceeding:

Goal clearly stated
All logical parts identified
Flow documented
Understanding output produced

进入下一步前需确认：

已明确表述目标
已识别所有逻辑模块
已梳理并记录流程
已生成理解输出内容

Reflection

反思

Did I understand the intent correctly?
Did I identify all logical parts? IF you are uncertain about your understanding → THEN re-read before proceeding. DO NOT guess.

我是否正确理解了Prompt的意图？
我是否识别了所有逻辑模块？若对理解内容不确定 → 则重新读取文件后再推进，禁止猜测。

FORBIDDEN

禁止操作

Proceeding without understanding the goal
Making changes based on assumptions
Skipping output format

未明确目标就推进
基于假设进行修改
跳过输出格式要求

ALLOWED

允许操作

Text output (understanding summary)
Re-reading file if needed

文本输出（理解总结）
必要时重新读取文件

On Failure

失败处理

IF intent unclear → THEN ask user for clarification
IF multiple interpretations → THEN present options and WAIT for user choice </understand_gate>

若意图不明确 → 则向用户请求澄清
若存在多种解读 → 则列出选项并等待用户选择 </understand_gate>

Step 3: RATE

步骤3：RATE

<rate_gate> STOP. DO NOT fix anything yet. Rate each logical part for issues first.

<rate_gate> STOP. 先不要修复问题，先对每个逻辑模块进行问题评级。

Pre-Conditions

前置条件

Step 2 (UNDERSTAND) completed
Understanding output produced

步骤2（UNDERSTAND）已完成
已生成理解输出内容

Issue Categories (MUST check all)

问题分类（必须全部检查）

Category	What to Look For	Severity
Weak Words	"consider", "might", "could", "may", "should" in critical sections	Critical
Missing Enforcement	Rules without FORBIDDEN/ALLOWED	High
Ambiguous Instructions	"do some", "handle", "process" without specifics	High
Referential Ambiguity	"it", "this", "that", "above", "below" without clear antecedent	High
Missing Output Format	Expected outputs without templates	Medium
Missing Gates	Phase transitions without checkpoints	Medium
Duplication	Same logic/rule repeated in multiple places (not just examples)	Medium
Verbose/Bloat	Sections >20 lines that could be tables; prose without constraints	Medium
Emoji as Instructions	Emojis used as commands instead of strong words	Medium
Redundancy	Same example repeated, unnecessary variations	Low
Low Density	Explanations that don't constrain behavior	Low

分类	检查要点	严重程度
弱词	关键部分出现"consider"、"might"、"could"、"may"、"should"	关键
缺乏强制约束	规则中未明确FORBIDDEN/ALLOWED	高
指令模糊	出现"do some"、"handle"、"process"等无具体说明的表述	高
指代模糊	出现"it"、"this"、"that"、"above"、"below"等无明确指代对象的表述	高
缺少输出格式	预期输出未提供模板	中
缺少检查门	阶段转换无检查点	中
重复内容	同一逻辑/规则在多处重复（示例除外）	中
冗余冗长	超过20行的章节可转换为表格；无约束的散文式内容	中
用表情符号当指令	使用表情符号代替明确指令	中
冗余内容	重复示例、不必要的变体	低
信息密度低	未约束行为的解释性内容	低

Rating Output (REQUIRED)

要求的评级输出

markdown

undefined

markdown

undefined

Issues Found

发现的问题

Part	Issue	Severity	Fix Needed
[Part name]	[Description]	Critical/High/Medium/Low	[What to do]

undefined

模块	问题描述	严重程度	修复方向
[模块名称]	[具体描述]	关键/高/中/低	修复方式

undefined

Gate Check

检查门验证

Verify before proceeding:

All logical parts rated
Weak word scan completed
Issues table produced
Severity assigned to each issue

进入下一步前需确认：

所有逻辑模块已完成评级
已完成弱词扫描
已生成问题表格
每个问题已分配严重程度

FORBIDDEN

禁止操作

Fixing issues before completing rating
Ignoring critical issues
Skipping weak word scan
Proceeding without issues table

完成评级前修复问题
忽略关键问题
跳过弱词扫描
未生成问题表格就推进

ALLOWED

允许操作

Text output (issues table)
Re-reading parts for rating

文本输出（问题表格）
为评级重新读取模块内容

On Failure

失败处理

IF no issues found → THEN MUST double-check with weak word scan
IF scan still clean → THEN document "No issues found" and proceed </rate_gate>

若未发现问题 → 则必须重新进行弱词扫描确认
若扫描后仍无问题 → 则记录"未发现问题"并进入下一步 </rate_gate>

Weak Word Reference

弱词参考表

Weak Word	Context	Replacement
consider, might, could, may	Critical section	MUST, REQUIRED
consider, might, could, may	Optional guidance	Remove or keep with "optionally"
should, prefer	Critical section	MUST
should, prefer	Soft guidance	Keep as-is
do some, handle, process	Any	Specify exact action: "Run X", "Call Y"
as needed, if necessary	Any	IF [condition] → THEN [action]
feel free to, you can	Required action	Remove entirely, use MUST
feel free to, you can	Optional action	"Optionally, you may..."

CRITICAL: Weak words in FORBIDDEN/MUST/NEVER sections MUST be replaced.

弱词	使用场景	替换词
consider, might, could, may	关键环节	MUST, REQUIRED
consider, might, could, may	可选引导	删除或添加"optionally"保留
should, prefer	关键环节	MUST
should, prefer	软性引导	保留原样
do some, handle, process	任何场景	指定具体操作："Run X", "Call Y"
as needed, if necessary	任何场景	IF [条件] → THEN [操作]
feel free to, you can	必要操作	完全删除，使用MUST
feel free to, you can	可选操作	"Optionally, you may..."

CRITICAL： MUST/FORBIDDEN/NEVER等关键规则中的弱词必须替换。

Step 4: FIX

步骤4：FIX

<fix_gate> STOP. Fix issues in priority order: Critical → High → Medium → Low.

<fix_gate> STOP. 按优先级修复问题：关键 → 高 → 中 → 低。

Pre-Conditions

前置条件

Step 3 (RATE) completed
Issues table produced

步骤3（RATE）已完成
已生成问题表格

Fix Priority (MUST follow order)

修复优先级（必须遵循）

Critical first - Weak words in MUST/FORBIDDEN contexts
High next - Missing enforcement, ambiguous instructions
Medium - Missing output formats, missing gates
Low last - Redundancy, density (only if value added)

先处理关键问题 - MUST/FORBIDDEN场景中的弱词
再处理高优先级问题 - 缺少强制约束、指令模糊
中优先级问题 - 缺少输出格式、缺少检查门
最后处理低优先级问题 - 冗余内容、信息密度（仅在有价值时处理）

Command Strength Hierarchy

指令强度层级

Strength	Keywords	Use For
Absolute	NEVER, ALWAYS, MUST, FORBIDDEN, CRITICAL	Non-negotiable rules
Stop	STOP, HALT, DO NOT proceed, WAIT	Gates/checkpoints
Required	REQUIRED, MANDATORY	Essential steps
Soft	should, prefer	Optional guidance only

强度	关键词	使用场景
绝对强制	NEVER, ALWAYS, MUST, FORBIDDEN, CRITICAL	不可协商的规则
强制停止	STOP, HALT, DO NOT proceed, WAIT	检查门/检查点
必须执行	REQUIRED, MANDATORY	必要步骤
软性引导	should, prefer	仅用于可选引导

Triple Lock Pattern (REQUIRED for Critical Rules)

三重锁定模式（关键规则必填）

1. STATE: "You MUST X"
2. FORBID: "FORBIDDEN: Not doing X"
3. REQUIRE: "REQUIRED: Verify X complete"

1. STATE: "You MUST X"
2. FORBID: "FORBIDDEN: Not doing X"
3. REQUIRE: "REQUIRED: Verify X complete"

Reasoning Block (REQUIRED Before Changes)

推理模块（修改前必填）

REQUIRED: Before making changes, produce a

<reasoning>

block:

markdown

<reasoning>
1. **Current state:** [What exists now]
2. **Goal:** [What we are trying to achieve]
3. **Approach:** [Why this specific change]
4. **Risk:** [What could go wrong]
</reasoning>

必须执行： 修改前需生成

<reasoning>

模块：

markdown

<reasoning>
1. **当前状态：** [现有内容]
2. **目标：** [要实现的效果]
3. **方案：** 选择该修改的原因
4. **风险：** 可能出现的问题
</reasoning>

Gate Template (When Adding Gates)

检查门模板（添加检查门时使用）

markdown

<[name]_gate>
**STOP. DO NOT proceed. [What to verify]**

markdown

<[名称]_gate>
**STOP. DO NOT proceed. [需验证的内容]**

Pre-Conditions

前置条件

[Previous step completed]

上一步已完成

Actions (REQUIRED)

必须执行的操作

[Action]

[操作内容]

Gate Check

检查门验证

Verify before proceeding:

[Condition]

进入下一步前需确认：

[验证条件]

FORBIDDEN

禁止操作

[What not to do]

[禁止的行为]

ALLOWED

允许操作

[What is permitted]

[允许的行为]

On Failure

失败处理

IF [condition] → THEN [recovery] </[name]_gate>

undefined

IF [条件] → THEN [恢复措施] </[名称]_gate>

undefined

Gate Check

检查门验证

Verify before proceeding:

All Critical issues fixed
All High issues fixed
Medium/Low addressed or documented as skipped
Reasoning block produced before changes

进入下一步前需确认：

所有关键问题已修复
所有高优先级问题已修复
中/低优先级问题已处理或记录为跳过
修改前已生成推理模块

FORBIDDEN

禁止操作

Over-strengthening soft guidance (keep "should" for optional items)
Changing logic that already works
Adding unnecessary complexity
Skipping Critical/High issues
Bloating: fixes that increase line count >10% without adding required gates

过度强化软性引导内容（可选内容保留"should"）
修改已正常运行的逻辑
添加不必要的复杂度
跳过关键/高优先级问题
冗余扩充：除非添加必要检查门，否则修复后行数增幅不得超过10%

ALLOWED

允许操作

Text output (draft fixes)
Iterating on fixes

文本输出（修复草稿）
迭代优化修复内容

On Failure

失败处理

IF over-strengthening detected → THEN revert and re-assess using RATE step criteria
IF unsure if logic changed → THEN compare before/after intent </fix_gate>

若发现过度强化 → 则撤销修改并重新按RATE步骤标准评估
若不确定是否修改了逻辑 → 则对比修改前后的意图 </fix_gate>

Step 5: VALIDATE

步骤5：VALIDATE

<validate_gate> STOP. DO NOT output yet. Validate all fixes against checklist.

<validate_gate> STOP. 先不要输出，根据检查清单验证所有修复内容。

Pre-Conditions

前置条件

Step 4 (FIX) completed
All Critical/High issues addressed

步骤4（FIX）已完成
所有关键/高优先级问题已处理

Validation Checklist (MUST complete all)

验证检查清单（必须全部完成）

Reflection (REQUIRED)

指代清晰度检查（必须完成）

MUST answer these questions:

Would I trust this prompt to execute reliably?
What's the weakest remaining section?
Did I change any original intent? (MUST be NO)

IF weakness identified → THEN fix or document as limitation IF intent changed → THEN STOP and revert. Return to UNDERSTAND step.

无模糊代词或位置指代（需明确指代对象）
所有实体名称统一
步骤/输出按名称引用，而非位置
所有交叉引用清晰明确
无隐含的"the"指代（需明确对象）
关键内容使用XML标签标注

Definition of Done (DoD) - Fast Final Gate

反思（必须完成）

ALL must be true before OUTPUT:

Single execution path (no ambiguous branches)
All inputs/outputs explicitly defined
All decision points use IF/THEN
No orphan references (every "it/this" resolved)

必须回答以下问题：

我是否信任该Prompt能可靠执行？
剩余最薄弱的环节是什么？
我是否修改了原始意图？（必须回答否）

若识别到薄弱环节 → 则修复或记录为限制项若已修改意图 → 则停止操作并撤销修改，返回UNDERSTAND步骤

Gate Check

完成标准（DoD）- 最终检查门

Verify before proceeding:

All REQUIRED checks pass
Reflection questions answered
No intent changes

输出前必须满足所有条件：

单一执行路径（无模糊分支）
所有输入/输出明确定义
所有决策点使用IF/THEN
无未明确指代的"it/this"等表述

FORBIDDEN

检查门验证

Outputting without completing validation
Skipping checklist items
Proceeding with failed checks

进入下一步前需确认：

所有必填检查项已通过
已回答反思问题
未修改原始意图

ALLOWED

禁止操作

Text output (validation results)
Returning to FIX step

未完成验证就输出
跳过检查清单项
未通过检查就推进

On Failure

允许操作

IF validation fails → THEN return to FIX step
IF intent changed → THEN return to UNDERSTAND step </validate_gate>

文本输出（验证结果）
返回FIX步骤重新处理

Step 6: OUTPUT

失败处理

<output_gate> STOP. Verify VALIDATE step passed before outputting.

若验证失败 → 则返回FIX步骤
若已修改意图 → 则返回UNDERSTAND步骤 </validate_gate>

Pre-Conditions

步骤6：OUTPUT

Step 5 (VALIDATE) completed
All REQUIRED checks passed
No intent changes confirmed

<output_gate> STOP. 确认VALIDATE步骤通过后再输出。

Output Format (REQUIRED - DO NOT deviate)

前置条件

markdown

undefined

步骤5（VALIDATE）已完成
所有必填检查项已通过
已确认未修改原始意图

Optimization Complete

要求的输出格式（必须严格遵循，不得偏离）

Summary

—

Issues Found: [N]
Fixes Applied: [N]
Intent Preserved: Yes

markdown

undefined

Changes Made

优化完成

—

总结

Category	Count	Examples
Command Strengthening	[N]	[Brief example]
Gates Added/Fixed	[N]	[Brief example]
Redundancy Removed	[N]	[Brief example]

发现问题数： [N]
已修复问题数： [N]
意图保留： 是

Optimized Document

已做修改

[Full optimized content]

undefined

分类	数量	示例
指令强化	[N]	[简要示例]
检查门添加/修复	[N]	[简要示例]
冗余内容移除	[N]	[简要示例]

FORBIDDEN

优化后的文档

Deviating from output format
Outputting without validation pass
Omitting the optimized document

[完整优化内容]

undefined

ALLOWED

禁止操作

StrReplace/Write to save optimized content
Text output (summary + document)

偏离输出格式
未通过验证就输出
省略优化后的文档

On Failure

允许操作

IF format deviates → THEN regenerate output
IF user requests changes → THEN return to FIX step </output_gate>

使用StrReplace/Write保存优化内容
文本输出（总结+文档）

Reference: Instruction Precedence

失败处理

<precedence_table> When rules conflict, follow this precedence (highest wins):

Priority	Category	Examples	Notes
1 (highest)	Safety/Tool Restrictions	FORBIDDEN tools, NEVER actions	Always wins
2	User explicit request	"I want X", "Do Y"	Overrides defaults
3	FORBIDDEN/MUST rules	"FORBIDDEN: changing logic"	Overrides preferences
4	Skill defaults	Default behaviors, templates	Baseline
5 (lowest)	Soft guidance	"prefer", "consider"	Yields to all above

Resolution rule: When two rules conflict, the higher priority wins. Document the conflict and resolution. </precedence_table>

若格式偏离 → 则重新生成输出内容
若用户要求修改 → 则返回FIX步骤 </output_gate>

Reference: Context Patterns

参考：指令优先级

<reasoning_patterns>

<precedence_table> 当规则冲突时，遵循以下优先级（优先级高的胜出）：

优先级	分类	示例	说明
1（最高）	安全/工具限制	禁止使用的工具、NEVER操作	始终优先
2	用户明确请求	"我想要X", "执行Y"	覆盖默认设置
3	FORBIDDEN/MUST规则	"FORBIDDEN: 修改逻辑"	覆盖偏好设置
4	Skill默认设置	默认行为、模板	基准规则
5（最低）	软性引导	"prefer", "consider"	服从所有更高优先级规则

冲突解决规则： 当两条规则冲突时，优先级高的规则胜出。需记录冲突及解决方式。 </precedence_table>

State Summaries (Context Retention)

参考：上下文模式

Use at the start of complex phases to prevent "forgetting":

markdown

<phase_start>
**State Summary:**
- **Goal:** [What are we solving?]
- **Progress:** [What is already done?]
- **Next:** [Immediate next step]
- **Blockers:** [Any issues?]
</phase_start>

REQUIRED: For multi-step fixes, produce a state summary every 3 fixes. </reasoning_patterns>

<reasoning_patterns>

Reference: High-Value vs Low-Value Content

状态总结（上下文保留）

<content_guide>

在复杂阶段开始时使用，防止"遗忘"：

markdown

<phase_start>
**状态总结：**
- **目标：** 我们要解决什么问题？
- **进度：** 已完成哪些内容？
- **下一步：** 立即要执行的步骤
- **障碍：** 存在哪些问题？
</phase_start>

必须执行： 对于多步骤修复，每完成3个修复需生成一次状态总结。 </reasoning_patterns>

REMOVE (Low Value)

参考：高价值 vs 低价值内容

Pattern	Action
Explanatory prose without actionable constraints	DELETE
Numeric scoring systems	Replace with pass/fail
Repeated examples	Keep 1-2 best
Long prose paragraphs	Convert to tables
Hedging language	Replace with MUST or DELETE
Emoji as instructions	Replace with strong command words (unless prompt output requires emoji)

<content_guide>

KEEP (High Value)

移除（低价值）

Pattern	Example
Tables with actions	`
Imperative verbs	STOP, EXECUTE, VERIFY
FORBIDDEN/ALLOWED lists	Clear capability control
IF/THEN conditionals	`IF X → THEN Y`
Gate checkpoints	STOP before transitions
</content_guide>

模式	操作
无操作约束的解释性散文	删除
数字评分系统	替换为通过/不通过
重复示例	保留1-2个最佳示例
长段落散文	转换为表格
模糊表述的弱词	替换为MUST或删除
用表情符号当指令	替换为强指令词（除非Prompt输出要求使用表情符号）

Quick Reference

保留（高价值）

Goal	Pattern
Force a stop	`STOP. DO NOT proceed.`
Require action	`REQUIRED: You MUST [action]`
Ban action	`FORBIDDEN: [action] until [condition]`
Force format	`OUTPUT FORMAT (REQUIRED): [template]`
Create checkpoint	`### Gate Check` with `- [ ]` items
Handle errors	`IF [error] → THEN [recovery]`
Critical rule	Triple Lock: STATE + FORBID + REQUIRE
Replace emoji	Emoji → strong command word (MUST, FORBIDDEN, etc.)

模式	示例
带操作的表格	`
命令式动词	STOP, EXECUTE, VERIFY
FORBIDDEN/ALLOWED列表	清晰的能力控制
IF/THEN条件语句	`IF X → THEN Y`
检查门验证点	阶段转换前的STOP检查
</content_guide>

Common Mistakes

快速参考

<common_mistakes>

Mistake	Why It Fails	Fix
Skipping UNDERSTAND step	Fixes wrong things, breaks working logic	ALWAYS produce Understanding output first
Fixing before RATE complete	Misses issues, inconsistent severity handling	Complete issues table before ANY fix
Over-strengthening soft guidance	"prefer" → "MUST" breaks optional flexibility	Keep "should/prefer" for truly optional items
Using "it/this/that"	Agent loses context, applies fix to wrong element	Name every entity explicitly
Outputting without VALIDATE	Weak words remain, conflicts undetected	MUST complete all checklist items
Changing working logic	User trusted original behavior	FORBIDDEN: If the logic works, don't touch it
Bloating the prompt	Verbose fixes overwhelm context, reduce density	MUST keep fixes concise; <10% line increase
</common_mistakes>

目标	模式
强制停止	`STOP. DO NOT proceed.`
要求执行操作	`REQUIRED: You MUST [action]`
禁止操作	`FORBIDDEN: [action] until [condition]`
强制格式	`OUTPUT FORMAT (REQUIRED): [template]`
创建检查点	`### Gate Check` 配合 `- [ ]` 项
错误处理	`IF [error] → THEN [recovery]`
关键规则	三重锁定：STATE + FORBID + REQUIRE
替换表情符号	表情符号 → 强指令词（MUST, FORBIDDEN等）

Before/After Example

常见错误

<before_after_example>

<common_mistakes>

错误	失败原因	修复方式
跳过UNDERSTAND步骤	修复错误内容，破坏正常逻辑	始终先生成理解输出内容
完成RATE前修复问题	遗漏问题，严重程度处理不一致	生成问题表格后再进行任何修复
过度强化软性引导	将"prefer"改为"MUST"破坏可选灵活性	真正的可选内容保留"should/prefer"
使用"it/this/that"	Agent丢失上下文，修复错误元素	明确命名所有实体
未验证就输出	仍存在弱词，未检测到冲突	必须完成所有检查清单项
修改正常运行的逻辑	用户信任原始行为	禁止操作：若逻辑正常，请勿修改
冗余扩充Prompt	冗长修复内容超出上下文，降低信息密度	必须保持修复简洁；行数增幅<10%
</common_mistakes>

Before (Weak)

前后对比示例

Consider checking the file before making changes.
You should validate the output.
Handle any errors that occur.
If needed, retry the operation.

<before_after_example>

After (Strong)

优化前（弱指令）

**REQUIRED:** You MUST read the file completely before making ANY changes.
**FORBIDDEN:** Outputting without validation.

Consider checking the file before making changes.
You should validate the output.
Handle any errors that occur.
If needed, retry the operation.

On Error

优化后（强指令）

IF file unreadable → THEN ask user for correct path
IF validation fails → THEN return to FIX step

**REQUIRED:** You MUST read the file completely before making ANY changes.
**FORBIDDEN:** Outputting without validation.

Retry Logic

错误处理

IF operation fails AND retry_count < 3 → THEN retry with exponential backoff
IF retry_count >= 3 → THEN STOP and report failure to user

undefined

IF file unreadable → THEN ask user for correct path
IF validation fails → THEN return to FIX step

Changes Applied

重试逻辑

Original	Issue	Fixed
"Consider checking"	Weak word in critical action	"REQUIRED: You MUST"
"should validate"	Weak enforcement	"FORBIDDEN: without validation"
"Handle any errors"	Ambiguous instruction	Explicit IF/THEN for each error
"If needed, retry"	Vague condition	Specific retry count + backoff
</before_after_example>

IF operation fails AND retry_count < 3 → THEN retry with exponential backoff
IF retry_count >= 3 → THEN STOP and report failure to user

undefined

—

已做修改

—

原始内容	问题	优化后
"Consider checking"	关键操作使用弱词	"REQUIRED: You MUST"
"should validate"	约束性弱	"FORBIDDEN: without validation"
"Handle any errors"	指令模糊	为每个错误明确IF/THEN规则
"If needed, retry"	条件模糊	明确重试次数 + 退避策略
</before_after_example>