X Extract Tweet with Comments - Content Production Guide

Collect posts and comments from Twitter/X around specific topics, organize netizen opinions and add host commentary, and generate narration scripts.

Warning: Core Principle - Document Output Oriented

The output of this skill serves subsequent stages, and all processes must be documented in files.

Do not just display in response - Results of each stage must be written to corresponding outputs/ files
Incremental saving - Write to files immediately after collecting/organizing each batch of content, don't wait until the end
Traceable - All content must be annotated with real sources (URLs) to ensure subsequent verification
Templates are for format reference only - Placeholders must be replaced with real content, copying example text is prohibited

Production Workflow

| Stage | Task | Output | Manual Intervention | |-------|------|--------|-------------------| | 1. Material Collection | Search/browse on X, incrementally record materials | 01_draft.jsonl | - | | 2. Extraction & Organization | Categorize opinions, discuss "my" viewpoint with user | 02_content_draft.md | - | | 3. Narration Generation | Generate formal narration script | 03_narration.md | Adjust narration content |

Stage 1: Material Collection

Output: outputs/01_draft.jsonl, incrementally record all materials.

Principle: Better more than less, missing materials cannot be traced later.

01_draft.jsonl Fields: type(tweet/comment), url, author, content, scroll_to_text, parent_url(required for comments)

Format example see templates/01_draft.jsonl

Warning: Critical Constraints (Must Follow)

Do not copy template example content - Templates are for format reference only, each record must be actually collected material
URLs must be real - Obtain from browser address bar or DOM, fabricating or using placeholders is strictly prohibited
Clear output file before starting collection - Ensure outputs/01_draft.jsonl doesn't contain old data or template content

Screenshot Usage Guidelines

Principle: Use screenshots sparingly, use recipes to extract content more.

| Purpose | Correct | Incorrect | |---------|---------|-----------| | Get content | x_extract_* recipes | Screenshot and let AI "read" | | Verify status | Screenshot check | - | | Backup position | Screenshot (comments may reorder) | - |

Stage 2: Extraction & Organization

Extract materials from 01_draft.jsonl, categorize and organize, discuss "my" opinions with user.

Output: outputs/02_content_draft.md, contains categorized opinions and "my comments".

Template see templates/02_content_draft.md

Query command: cat outputs/01_draft.jsonl | jq 'select(.type=="tweet")'

Stage 3: Narration Generation

Generate formal narration script based on 02_content_draft.md.

Output: outputs/03_narration.md

Template see templates/03_narration.md

Style Requirements:

Conversational, brisk language rhythm
Get straight to the topic, no self-introduction
Start with an impactful hook to grab attention

Manual Operation: Adjust narration content to ensure natural and smooth expression.

Common Issues

| Issue | Solution | |-------|----------| | Cannot find target element | Use longer text snippets | | Comment position changes | Screenshot backup during collection |

Reference Documentation

Twitter element features and selectors: REFERENCE.md
frago CDP commands: uv run frago --help