Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/yuuka-dev/xpostdeleter

削除めんどうだよね？そのためのやつ
https://github.com/yuuka-dev/xpostdeleter

Last synced: 2 months ago
JSON representation

削除めんどうだよね？そのためのやつ

Awesome Lists containing this project

README 

          
# XPostDeleter



**[English](#english) | [Bahasa Indonesia](#bahasa-indonesia)**



---





## English



### What is this?



XPostDeleter scans your X (Twitter) archive and automatically deletes posts that contain **your face**, **NSFW content**, or **specific keywords** — your personal "black history" eraser.



**How it works:**

1. `archive_scanner.py` — reads your X archive, detects faces/NSFW with AI and/or scans text with LLM, outputs `delete_hit_list.csv`

2. `delete_agent.py` — reads the CSV and deletes each post via Selenium with human-like mouse movement to avoid bot detection



### Features



- Face detection with [InsightFace](https://github.com/deepinsight/insightface) — matches your face against reference selfies

- NSFW detection with [NudeNet](https://github.com/notAI-tech/NudeNet)

- 3-stage text analysis (`--text`): keyword match → Gemini → Claude

- Human-like mouse movement (Bézier curve + easing + jitter) to reduce bot detection risk

- Checkpoint saving — progress is written to CSV after every deletion, safe to resume

- RT (repost) handling — reposts are separated to `rt_hit_list.csv` and processed separately

- Progress display with severity filtering



### Requirements



- Python 3.11+

- Google Chrome + matching [ChromeDriver](https://googlechromelabs.github.io/chrome-for-testing/)

- GPU recommended (CUDA) for InsightFace — CPU works but is slow

- Gemini API key (free tier is sufficient) for text analysis



### Setup



#### 1. Clone & create virtual environment



```bash

git clone https://github.com/yourname/XPostDeleter.git

cd XPostDeleter

python -m venv .venv



# Windows

.venv\Scripts\activate



# macOS / Linux

source .venv/bin/activate

```



#### 2. Install dependencies



```bash

pip install -r requirements.txt

```



#### 3. Configure `.env`



```bash

cp .env.example .env

```



Edit `.env`:



```env

# Chrome user data directory

USER_DATA_DIR=C:\Users\YourName\AppData\Local\Google\Chrome\User Data



# Required only for --text mode

GEMINI_API_KEY=your_gemini_api_key_here



# Optional — used for borderline cases escalated from Gemini

ANTHROPIC_API_KEY=your_anthropic_api_key_here

```



#### 4. Add reference selfies



Place **2–5 clear selfies of yourself** in `reference_media/`. These are used by InsightFace to recognize your face in archive images.



```

reference_media/

  selfie1.jpg

  selfie2.jpg

```



> `reference_media/` is in `.gitignore` — your photos will never be committed.



#### 5. (Optional) Configure keywords for text analysis



```bash

cp keywords.json.example keywords.json

```



Edit `keywords.json` to add the words you want to detect in tweet text. Three severity tiers are supported:



```json

{

  "high":   ["word that must go"],

  "medium": ["word that should go"],

  "low":    ["word to consider"]

}

```



> `keywords.json` is in `.gitignore` — your personal keyword list will never be committed.



---



### Step 1 — Prepare your X Archive



1. Go to **X Settings → Your account → Download an archive of your data**

2. Wait for the email, download and extract the zip

3. Place the extracted folder as `XArchive/` in the project root:



```

XPostDeleter/

  XArchive/

    data/

      tweets.js

      tweets_media/

        12345678-photo.jpg

        ...

```



---



### Step 2 — Generate the delete list



#### Image / video scan only (default)



Detects posts with your face or NSFW content:



```bash

python archive_scanner.py

```



> If `delete_hit_list.csv` already exists, image/video analysis is automatically skipped — only new tweets are processed.



#### Image + text analysis



Also scans all tweet text with keyword matching and LLM judgment:



```bash

python archive_scanner.py --text

```



**Text analysis pipeline (`--text`):**



| Stage | Engine | When used |

|-------|--------|-----------|

| 1 — Keyword | local `keywords.json` | always (instant, free) |

| 2 — Gemini | Gemini API (free tier) | tweets that pass keyword filter |

| 3 — Claude | Claude API (optional) | tweets Gemini rates as uncertain |



> The Gemini free tier (15 RPM) is sufficient. A 10,000-tweet archive takes roughly 30 minutes for text analysis.



This creates (or updates) `delete_hit_list.csv`:



| Column | Description |

|--------|-------------|

| `created_at` | Post timestamp |

| `delete_url` | URL of the post |

| `severity` | Risk score 0–5 |

| `risk_tags` | e.g. `FACE_SELFIE`, `NSFW_HIGH`, `TEXT_KEYWORD`, `TEXT_GEMINI` |

| `full_text` | Post text |

| `hapus` | Deletion status (empty = pending, `sudah` = done) |



**Always review the CSV and remove any rows you want to keep before proceeding.**



---



### Step 3 — Run the delete agent



#### Option A — Attach to an already-running Chrome



1. Close all Chrome windows, then launch Chrome with remote debugging:



```powershell

& "C:\Program Files\Google\Chrome\Application\chrome.exe" `

  --remote-debugging-port=9222 `

  --user-data-dir="$env:USER_DATA_DIR" `

  --profile-directory="Default"

```



2. Log in to X manually in that Chrome window.



3. Run the agent:



```bash

python delete_agent.py --attach-existing --limit 50

```



#### Option B — Let Selenium launch Chrome automatically



```bash

python delete_agent.py --limit 50

```



#### Key options



| Option | Default | Description |

|--------|---------|-------------|

| `--limit N` | 50 | Max posts to delete per run |

| `--allow-low-severity` | off | Also delete severity < 1 posts |

| `--attach-existing` | off | Attach to running Chrome on port 9222 |

| `--rt` | off | Process `rt_hit_list.csv` (unretweet mode) |

| `--log-file PATH` | `xpostdeleter.log` | Log file path |

| `--debug-rt` | off | Verbose RT/repost detection logging |



---



### Project structure



```

XPostDeleter/

  delete_agent.py        # Entry point (CLI)

  actions.py             # Delete / unretweet routines

  browser.py             # ChromeDriver factory & navigation

  human.py               # Human-like mouse, scroll, timing

  utils.py               # Logging tee & progress bar

  archive_scanner.py     # Archive scan → delete_hit_list.csv

  image_analyzer.py      # Face recognition & NSFW detection

  text_analyzer.py       # 3-stage text analysis (keyword / Gemini / Claude)

  keywords.json.example  # Keyword list template (copy to keywords.json)

  .env.example           # Environment variable template

  requirements.txt       # Python dependencies

```



### Disclaimer



> **Read carefully before use.**



- This tool is for **personal use only** — deleting your own posts.

- Automated browser interaction may violate X's Terms of Service — **use at your own risk**.

- During execution, **do not touch the mouse** (pyautogui controls it).

- Keep batch sizes reasonable (20–50 per run) to reduce detection risk.

- **AI detection (image, NSFW, and text) is not perfect.** False positives will occur — posts that have nothing to do with your intended targets may end up on the delete list. **Always review `delete_hit_list.csv` before running the delete agent.** The authors take no responsibility for any posts deleted as a result of false positives or misdetection.



---

---





## Bahasa Indonesia



### Apa ini?



XPostDeleter memindai arsip X (Twitter) kamu dan secara otomatis menghapus postingan yang mengandung **wajahmu**, **konten NSFW**, atau **kata kunci tertentu** — penghapus "jejak digital memalukan" pribadimu.



**Cara kerjanya:**

1. `archive_scanner.py` — membaca arsip X, mendeteksi wajah/NSFW dengan AI dan/atau memindai teks dengan LLM, menghasilkan `delete_hit_list.csv`

2. `delete_agent.py` — membaca CSV dan menghapus setiap postingan lewat Selenium dengan gerakan mouse seperti manusia untuk menghindari deteksi bot



### Fitur



- Deteksi wajah dengan [InsightFace](https://github.com/deepinsight/insightface) — mencocokkan wajahmu dengan foto referensi selfie

- Deteksi NSFW dengan [NudeNet](https://github.com/notAI-tech/NudeNet)

- Analisis teks 3 tahap (`--text`): pencocokan keyword → Gemini → Claude

- Gerakan mouse seperti manusia (kurva Bézier + easing + jitter) untuk mengurangi risiko deteksi bot

- Penyimpanan checkpoint — progres ditulis ke CSV setelah setiap penghapusan, aman untuk dilanjutkan

- Penanganan RT (repost) — repost dipisah ke `rt_hit_list.csv` dan diproses terpisah

- Tampilan progres dengan filter tingkat keparahan



### Persyaratan



- Python 3.11+

- Google Chrome + [ChromeDriver](https://googlechromelabs.github.io/chrome-for-testing/) yang sesuai

- GPU direkomendasikan (CUDA) untuk InsightFace — CPU bisa tapi lambat

- Gemini API key (tier gratis sudah cukup) untuk analisis teks



### Persiapan



#### 1. Clone & buat virtual environment



```bash

git clone https://github.com/yourname/XPostDeleter.git

cd XPostDeleter

python -m venv .venv



# Windows

.venv\Scripts\activate



# macOS / Linux

source .venv/bin/activate

```



#### 2. Install dependensi



```bash

pip install -r requirements.txt

```



#### 3. Konfigurasi `.env`



```bash

cp .env.example .env

```



Edit `.env`:



```env

# Direktori data pengguna Chrome

USER_DATA_DIR=C:\Users\NamaKamu\AppData\Local\Google\Chrome\User Data



# Wajib hanya untuk mode --text

GEMINI_API_KEY=your_gemini_api_key_here



# Opsional — digunakan untuk kasus meragukan yang dieskalasi dari Gemini

ANTHROPIC_API_KEY=your_anthropic_api_key_here

```



#### 4. Tambahkan foto selfie referensi



Taruh **2–5 selfie yang jelas dari dirimu** di folder `reference_media/`. Ini digunakan InsightFace untuk mengenali wajahmu di foto-foto arsip.



```

reference_media/

  selfie1.jpg

  selfie2.jpg

```



> `reference_media/` ada di `.gitignore` — foto-foto kamu tidak akan pernah ikut ter-commit.



#### 5. (Opsional) Konfigurasi keyword untuk analisis teks



```bash

cp keywords.json.example keywords.json

```



Edit `keywords.json` untuk menambahkan kata-kata yang ingin kamu deteksi di teks tweet. Tersedia tiga tingkat keparahan:



```json

{

  "high":   ["kata yang harus dihapus"],

  "medium": ["kata yang sebaiknya dihapus"],

  "low":    ["kata yang perlu dipertimbangkan"]

}

```



> `keywords.json` ada di `.gitignore` — daftar keyword pribadimu tidak akan pernah ikut ter-commit.



---



### Langkah 1 — Siapkan Arsip X



1. Buka **X Settings → Your account → Download an archive of your data**

2. Tunggu email, unduh dan ekstrak zip-nya

3. Taruh folder hasil ekstrak sebagai `XArchive/` di root proyek:



```

XPostDeleter/

  XArchive/

    data/

      tweets.js

      tweets_media/

        12345678-photo.jpg

        ...

```



---



### Langkah 2 — Buat daftar hapus



#### Pindai gambar / video saja (default)



Mendeteksi postingan yang mengandung wajahmu atau konten NSFW:



```bash

python archive_scanner.py

```



> Jika `delete_hit_list.csv` sudah ada, analisis gambar/video otomatis dilewati — hanya tweet baru yang diproses.



#### Gambar + analisis teks



Juga memindai seluruh teks tweet dengan pencocokan keyword dan penilaian LLM:



```bash

python archive_scanner.py --text

```



**Pipeline analisis teks (`--text`):**



| Tahap | Engine | Kapan digunakan |

|-------|--------|-----------------|

| 1 — Keyword | `keywords.json` lokal | selalu (instan, gratis) |

| 2 — Gemini | Gemini API (tier gratis) | tweet yang lolos filter keyword |

| 3 — Claude | Claude API (opsional) | tweet yang dinilai meragukan oleh Gemini |



> Tier gratis Gemini (15 RPM) sudah cukup. Arsip 10.000 tweet membutuhkan sekitar 30 menit untuk analisis teks.



Ini akan membuat (atau memperbarui) `delete_hit_list.csv`:



| Kolom | Keterangan |

|-------|------------|

| `created_at` | Waktu postingan |

| `delete_url` | URL postingan |

| `severity` | Skor risiko 0–5 |

| `risk_tags` | misal `FACE_SELFIE`, `NSFW_HIGH`, `TEXT_KEYWORD`, `TEXT_GEMINI` |

| `full_text` | Teks postingan |

| `hapus` | Status penghapusan (kosong = belum, `sudah` = selesai) |



**Selalu tinjau CSV dan hapus baris yang ingin kamu pertahankan sebelum melanjutkan.**



---



### Langkah 3 — Jalankan agen penghapus



#### Opsi A — Menempel ke Chrome yang sudah berjalan



1. Tutup semua jendela Chrome, lalu jalankan Chrome dengan remote debugging:



```powershell

& "C:\Program Files\Google\Chrome\Application\chrome.exe" `

  --remote-debugging-port=9222 `

  --user-data-dir="$env:USER_DATA_DIR" `

  --profile-directory="Default"

```



2. Login ke X secara manual di jendela Chrome tersebut.



3. Jalankan agennya:



```bash

python delete_agent.py --attach-existing --limit 50

```



#### Opsi B — Biarkan Selenium menjalankan Chrome secara otomatis



```bash

python delete_agent.py --limit 50

```



#### Opsi-opsi utama



| Opsi | Default | Keterangan |

|------|---------|------------|

| `--limit N` | 50 | Maksimal postingan yang dihapus per run |

| `--allow-low-severity` | mati | Juga hapus postingan severity < 1 |

| `--attach-existing` | mati | Menempel ke Chrome yang berjalan di port 9222 |

| `--rt` | mati | Proses `rt_hit_list.csv` (mode batalkan repost) |

| `--log-file PATH` | `xpostdeleter.log` | Path file log |

| `--debug-rt` | mati | Log deteksi RT/repost yang lebih detail |



---



### Struktur proyek



```

XPostDeleter/

  delete_agent.py        # Entry point (CLI)

  actions.py             # Rutinitas hapus / batalkan repost

  browser.py             # Factory ChromeDriver & navigasi

  human.py               # Mouse, scroll, timing seperti manusia

  utils.py               # Logging tee & progress bar

  archive_scanner.py     # Pindai arsip → delete_hit_list.csv

  image_analyzer.py      # Pengenalan wajah & deteksi NSFW

  text_analyzer.py       # Analisis teks 3 tahap (keyword / Gemini / Claude)

  keywords.json.example  # Template daftar keyword (salin ke keywords.json)

  .env.example           # Template variabel lingkungan

  requirements.txt       # Dependensi Python

```



### Disclaimer



> **Baca dengan seksama sebelum digunakan.**



- Alat ini hanya untuk **penggunaan pribadi** — menghapus postingan milikmu sendiri.

- Interaksi browser otomatis mungkin melanggar Ketentuan Layanan X — **gunakan dengan risiko sendiri**.

- Saat dijalankan, **jangan sentuh mouse** (pyautogui mengontrolnya).

- Jaga ukuran batch tetap wajar (20–50 per run) untuk mengurangi risiko deteksi.

- **Deteksi AI (gambar, NSFW, dan teks) tidak sempurna.** False positive akan terjadi — postingan yang sama sekali tidak terkait dengan target yang kamu maksud bisa saja masuk ke daftar hapus. **Selalu tinjau `delete_hit_list.csv` sebelum menjalankan agen penghapus.** Penulis tidak bertanggung jawab atas postingan yang terhapus akibat false positive atau kesalahan deteksi.