Skip to content

dean9703111/ai-agent-skill-for-video-workflow / .agent/skills/srt-enhancer

SRT Enhancer

Verified

This skill should be used when the user asks to "enhance srt subtitles", "optimize srt with original script", "fix srt typos from markdown", "compare srt with origin.md", "improve subtitle accuracy", "add spaces to srt", "優化 srt 逐字稿", "比對原稿修正字幕", or needs to refine SRT subtitle files by comparing them with an original markdown reference document.

FileKindSize
SKILL.mdskill md8.2 KB
references/enhancement-rules.mdmarkdown8.5 KB
references/spacing-examples.mdmarkdown8.5 KB

references/enhancement-rules.md

View on GitHub
# Enhancement Rules

This document provides detailed rules for enhancing SRT subtitles by comparing them with `origin.md` reference documents.

## Semantic Matching Strategies

### Understanding Context

When matching SRT content with `origin.md`:

1. **Conceptual Matching**: Match based on meaning, not exact words
   - SRT: "今天我們來聊聊機器學習"
   - origin.md: "今天要討論機器學習的概念"
   - Match: ✓ (same concept, different wording)

2. **Sequential Context**: Use surrounding subtitles for context
   - Consider previous and next subtitle segments
   - Identify topic flow and narrative structure
   - Match based on position in content flow

3. **Keyword Anchoring**: Use distinctive terms as anchors
   - Technical terms (Python, API, Docker)
   - Proper nouns (人名, 地名, 產品名)
   - Numbers and statistics
   - Unique phrases

4. **Fuzzy Matching**: Account for transcription variations
   - Homophones in Chinese (的/得/地)
   - Similar-sounding words
   - Abbreviated vs. full forms
   - Colloquial vs. formal language

### Matching Algorithm

For each subtitle segment:

1. Extract key terms (proper nouns, technical terms, numbers)
2. Search for these terms in `origin.md`
3. Identify candidate matches within reasonable proximity
4. Compare context and meaning
5. Select best match based on:
   - Keyword overlap
   - Semantic similarity
   - Sequential position
   - Context alignment

## Typo Detection and Correction

### Common Transcription Errors

1. **Homophone Errors**:
   - 的/得/地 confusion
   - 在/再 confusion
   - 做/作 confusion
   - 他/她/它 confusion

2. **Similar Character Errors**:
   - 工具/工俱
   - 功能/公能
   - 設定/設訂
   - 資料/資料

3. **Missing Characters**:
   - Dropped particles (的, 了, 著)
   - Incomplete words
   - Truncated phrases

4. **Extra Characters**:
   - Duplicate words
   - Filler words (那個, 就是, 然後)
   - Transcription artifacts

### Correction Strategy

1. **Identify Discrepancy**: Compare SRT text with matched `origin.md` content
2. **Verify Error**: Confirm it's a typo, not intentional variation
3. **Apply Correction**: Replace with reference version
4. **Validate Context**: Ensure correction makes sense in subtitle context

### Conservative Approach

When uncertain:
- Preserve original text
- Don't force corrections
- Only correct clear errors
- Maintain speaker's voice and style

## Proper Noun Extraction

### Categories of Proper Nouns

1. **Personal Names**:
   - Chinese names: 李明, 王小華
   - English names: Steve Jobs, Bill Gates
   - Nicknames and aliases

2. **Company/Organization Names**:
   - Tech companies: Google, Microsoft, Apple
   - Organizations: WHO, UNESCO
   - Brands: iPhone, Windows

3. **Product Names**:
   - Software: Python, Docker, Kubernetes
   - Hardware: MacBook, ThinkPad
   - Services: GitHub, AWS, Azure

4. **Technical Terms**:
   - Programming languages: JavaScript, TypeScript
   - Frameworks: React, Vue, Django
   - Concepts: API, REST, GraphQL

5. **Place Names**:
   - Cities: 台北, 上海, New York
   - Countries: 美國, 日本, Singapore
   - Landmarks: 101, 故宮

### Extraction Process

From `origin.md`:

1. **Identify Proper Nouns**:
   - Scan for capitalized English words
   - Identify Chinese proper nouns by context
   - Extract technical terminology
   - Note product and brand names

2. **Build Reference Dictionary**:
   - Create mapping of proper nouns
   - Include variations and abbreviations
   - Note correct capitalization
   - Record context for disambiguation

3. **Apply to SRT**:
   - Match proper nouns in subtitles
   - Standardize to reference version
   - Correct capitalization
   - Fix spacing around proper nouns

### Standardization Rules

1. **Capitalization**:
   - Follow reference document exactly
   - Maintain brand-specific styling (iPhone, macOS, GitHub)
   - Preserve acronym capitalization (API, REST, HTTP)

2. **Spacing**:
   - Add spaces around English proper nouns in Chinese text
   - Maintain internal spacing in multi-word names
   - Follow brand guidelines (e.g., "App Store" not "AppStore")

3. **Consistency**:
   - Use same form throughout (不要 Python/python 混用)
   - Standardize abbreviations (API vs. A.P.I.)
   - Maintain terminology consistency

## Terminology Consistency

### Building Terminology Database

From `origin.md`, extract:

1. **Domain-Specific Terms**:
   - Technical vocabulary
   - Industry jargon
   - Specialized concepts

2. **Preferred Translations**:
   - English → Chinese mappings
   - Consistent translation choices
   - Standard terminology

3. **Abbreviations and Acronyms**:
   - Full form vs. abbreviated form
   - When to use which
   - Correct formatting

### Application Strategy

1. **First Occurrence**: Establish terminology from reference
2. **Subsequent Uses**: Apply consistently throughout SRT
3. **Context Sensitivity**: Adjust for spoken vs. written context
4. **Preserve Meaning**: Don't force terminology that changes meaning

## Edge Cases and Special Handling

### Multiple Matches

When SRT content matches multiple sections in `origin.md`:
- Use sequential context to determine correct match
- Consider timestamp position in video
- Match based on topic flow
- Choose most semantically similar match

### No Match Found

When SRT content has no clear match in `origin.md`:
- Still apply spacing rules
- Don't invent corrections
- Preserve original text
- Flag for manual review if critical

### Paraphrasing

When SRT paraphrases `origin.md`:
- Recognize semantic equivalence
- Don't force exact wording
- Correct only clear errors
- Preserve speaker's phrasing style

### Colloquial Language

When SRT uses colloquial speech:
- Respect spoken language style
- Don't over-formalize
- Correct typos, not style
- Maintain natural flow

### Filler Words

Common in spoken transcripts:
- 那個, 就是, 然後, 嗯, 啊
- Generally preserve unless clearly errors
- Remove only if reference omits them
- Maintain natural speech patterns

## Quality Assurance

### Pre-Enhancement Checks

- [ ] `origin.md` exists and is readable
- [ ] SRT file is valid and parseable
- [ ] Both files use consistent encoding (UTF-8)

### During Enhancement

- [ ] Each correction has reference basis
- [ ] Timestamps remain unchanged
- [ ] Subtitle numbering preserved
- [ ] No content added from markdown
- [ ] Spacing rules applied consistently

### Post-Enhancement Validation

- [ ] Output file is `enhanced.srt`
- [ ] Subtitle count matches original
- [ ] All timestamps identical to original
- [ ] Proper nouns standardized
- [ ] English/numbers have proper spacing
- [ ] No semantic meaning changed
- [ ] SRT format is valid

## Examples

### Example 1: Typo Correction

**origin.md**: "這個工具非常強大"
**SRT Original**: "這個工俱非常強大"
**SRT Enhanced**: "這個工具非常強大"
**Reason**: 工俱 → 工具 (typo correction)

### Example 2: Proper Noun Standardization

**origin.md**: "使用 Python 3.9 開發"
**SRT Original**: "使用python3.9開發"
**SRT Enhanced**: "使用 Python 3.9 開發"
**Reason**: Capitalization + spacing

### Example 3: Terminology Consistency

**origin.md**: "機器學習模型"
**SRT Original**: "機器學習模形"
**SRT Enhanced**: "機器學習模型"
**Reason**: 模形 → 模型 (correct terminology)

### Example 4: No Change Needed

**origin.md**: "今天要討論這個主題"
**SRT Original**: "今天我們來聊這個主題"
**SRT Enhanced**: "今天我們來聊這個主題"
**Reason**: Paraphrasing is acceptable, no errors

### Example 5: Spacing Only

**origin.md**: "使用 API 進行整合"
**SRT Original**: "使用API進行整合"
**SRT Enhanced**: "使用 API 進行整合"
**Reason**: Add spacing around English acronym

## Best Practices

1. **Be Conservative**: When in doubt, preserve original
2. **Context is Key**: Use surrounding content for validation
3. **Respect Speech Patterns**: Don't over-formalize spoken language
4. **Verify Changes**: Each correction should have clear justification
5. **Maintain Flow**: Ensure corrections don't disrupt reading rhythm
6. **Test Output**: Validate enhanced SRT is still valid and readable

## Common Pitfalls to Avoid

1. **Over-Correction**: Changing style instead of errors
2. **Adding Content**: Including markdown content not in SRT
3. **Timestamp Modification**: Never change timing
4. **Forced Matching**: Matching unrelated content
5. **Ignoring Context**: Correcting based on isolated words
6. **Inconsistent Application**: Applying rules selectively

Focus on improving accuracy while preserving the natural flow and structure of the original subtitles.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall