Codex CLI Complete Guide

Sora 2 Prompt Version Control Implementation: Team Operations & Credit Optimization

This is a follow-up to the morning article

Morning article: Sora 2 Prompt Playbook

Goals

• Automatically record prompt change history with GitHub Gist and extract improvement patterns from diffs
• Build a team-wide prompt database with Notion API for instant search and reuse of successful examples
• Reduce wasted credit consumption by 20%+ through automated diff review

Architecture Overview

[Sora 2 CLI/Web] → (1) GitHub Gist API
                          ↓ (Save as JSON)
                    (2) Notion Database
                          ↓ (Tag search & diff display)
                    (3) Diff Visualization (gh/diffutils)
                          ↓
                    [Review & Improvement]

1. GitHub Gist: Save prompts in JSON format, record change reasons in commit logs
2. Notion Database: Centrally manage tags, success/failure flags, and credit consumption
3. Diff Tools: Visualize modification patterns using gh gist view --diff or git diff

Implementation Steps

Step 1: Prompt History Management with GitHub Gist

Use GitHub Gist to place each prompt under independent version control.

# Create Gist (initial)
gh gist create sora_prompt_v1.json --public

# Update (with comment)
gh gist edit <gist_id> --add sora_prompt_v2.json \
  --description "Fix: Changed camera work to 'gentle handheld' → viewability +15%"

JSON Format Example:

{
  "id": "001",
  "prompt": "A 50-year-old fisherman prepares for fishing at dawn port...",
  "metadata": {
    "resolution": "1080p",
    "length": "12s",
    "credits_used": 45,
    "success": true,
    "tags": ["documentary", "natural-light"]
  }
}

• Benefits: Track change reasons in commit logs, easy team sharing
• Note: Don't consolidate massive prompts into one file; split by scene

Step 2: Automated Recording to Notion DB

Batch-register prompts and metadata using Notion API to build a searchable database.

import requests
import json

NOTION_TOKEN = "secret_xxx"
DATABASE_ID = "abc123"

def add_prompt_to_notion(prompt_data):
    url = f"https://api.notion.com/v1/pages"
    headers = {
        "Authorization": f"Bearer {NOTION_TOKEN}",
        "Content-Type": "application/json",
        "Notion-Version": "2022-06-28"
    }
    payload = {
        "parent": {"database_id": DATABASE_ID},
        "properties": {
            "Name": {"title": [{"text": {"content": prompt_data["id"]}}]},
            "Prompt": {"rich_text": [{"text": {"content": prompt_data["prompt"]}}]},
            "Credits": {"number": prompt_data["metadata"]["credits_used"]},
            "Success": {"checkbox": prompt_data["metadata"]["success"]},
            "Tags": {"multi_select": [{"name": t} for t in prompt_data["metadata"]["tags"]]}
        }
    }
    response = requests.post(url, headers=headers, json=payload)
    return response.json()

# Usage example
with open("sora_prompt_v2.json") as f:
    data = json.load(f)
    add_prompt_to_notion(data)

• Benefits: Instant filtering by tags, credit consumption aggregation
• Note: Consider Notion API rate limits (3req/sec); batch processing recommended

Step 3: Automated Diff Review

Visualize modifications using GitHub CLI's diff features.

# Display Gist diff (last 2 versions)
gh gist view <gist_id> --files sora_prompt_v1.json,sora_prompt_v2.json | diff -u - -

# Detailed comparison using git diff
git diff --no-index sora_prompt_v1.json sora_prompt_v2.json --word-diff=color

Output Example:

-"Action": "handheld camera, violent shake"
+"Action": "handheld camera, gentle shake"

• Automation Idea: Notify diff to Slack on commit via GitHub Actions
• Advanced Use: Analyze correlation between diff line count and credit reduction → target maximum effect with minimal changes

Benchmark: Comparison of Version Control Methods

Method	Setup Time	Team Sharing	Diff Tracking	Credit Reduction Effect
Text File + Google Drive	5 min	△ (manual sync)	×	0%
GitHub Gist	10 min	◎	◎	15%
Gist + Notion API	30 min	◎	◎	20-25%

Measurement Conditions: 3-person team, 2 weeks, 120 prompts total. Counted cases where duplicate attempts were avoided through diff review.

Failure Patterns and Countermeasures

Symptom	Cause	Countermeasure
Gists scattered and unsearchable	Lack of file naming conventions	Unify to {project}_{scene}_{version}.json format
Notion sync delays / duplicate registration	Insufficient API retry handling	Implement requests.Session() + exponential backoff
Too many diffs to read	Changing multiple elements at once	Enforce 1 commit = 1 element change rule
Credit reduction not tangible	Missing baseline measurement	Week 1: logging only, Week 2+: compare improvement effects

Automation & Extension Ideas

• GitHub Actions Integration: Auto-register to Notion DB on commit, insert diff into PR comments
• Slack Bot: Retrieve similar prompts from Notion DB with /sora search #documentary
• Cost Prediction API: Pre-display "This prompt estimated at 50 credits" from historical data
• A/B Test Support: Generate 2 versions of same scene in parallel, compare viewership & credit efficiency
• Metadata Expansion: Add generation timestamp, assignee, feedback comments to accelerate team learning

Next Steps

• Sora 2 Audio Engineering Guide - Integrated management of audio and prompts
• GitHub CLI Complete Guide - Detailed Gist operation commands
• Notion API Reference - Official documentation for Database operations