misc 0 0 0

Gearsoldier0731/gearz-leakfinder

published by @Gearsoldier0731

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
1.7K
Last 30 days
98
Likes
-
Size
5.3 MB
Created on HF
2025-08-18
Age
1 years ago

Description

πŸ”Ž GEARZ LeakFinder

Redaction-first OSINT hunting tool β€” scan code repos, package registries, Docker images, CI/CD logs, and archived web captures for accidentally leaked secrets.Built for bug bounty pros, red teamers, and learners who want to see how real leaks are uncovered (safely, with redaction).

✨ Features

Adapters: GitHub β€’ Wayback Machine β€’ DockerHub β€’ npm β€’ PyPI β€’ CI/CD logs

Scope Builder UI: point-and-click to define your targets β†’ auto-saves to… See the full description on the dataset page: https://huggingface.co/datasets/Gearsoldier0731/gearz-leakfinder.

Tags

license:mitregion:usosintsecurityleak-detectiongithubdockerhubnpmpypinextjsapp

README current version from Hugging Face


pretty_name: GEARZ LeakFinder
license: mit
tags:

  • osint
  • security
  • leak-detection
  • github
  • dockerhub
  • npm
  • pypi
  • nextjs
  • app

πŸ”Ž GEARZ LeakFinder

GEARZ LeakFinder Theme

Redaction-first OSINT hunting tool β€” scan code repos, package registries, Docker images, CI/CD logs, and archived web captures for accidentally leaked secrets.
Built for bug bounty pros, red teamers, and learners who want to see how real leaks are uncovered (safely, with redaction).


✨ Features

  • Adapters: GitHub β€’ Wayback Machine β€’ DockerHub β€’ npm β€’ PyPI β€’ CI/CD logs
  • Scope Builder UI: point-and-click to define your targets β†’ auto-saves to scope.yaml
  • Redaction by default: sensitive values are masked before logging
  • Tutor Mode: step-by-step explanations of what’s happening
  • AI Summaries: results are explained in plain language by local or hosted LLMs
  • Export Pack: one-click download of JSONL findings, AI summary, and Merkle audit
  • Live Logs: stream engine output directly in the UI
  • Theme toggle: dark, dim, or sunburst modes

πŸ› οΈ Setup

  1. Clone and install
    git clone https://github.com/Gearsoldier/gearz-leakfinder.git
    cd gearz-leakfinder
    npm install
    Auth tokens
    

GitHub: Generate a Personal Access Token (classic, public_repo scope is enough).

Hugging Face: HF_TOKEN for AI summaries.

Export tokens into your shell (add to ~/.bashrc for permanence):

export GITHUB_TOKEN=ghp_xxxxx
export HF_TOKEN=hf_xxxxx
Start UI

npm run dev
Open β†’ http://localhost:3000

🎯 Usage

  1. Build your scope
    From the UI:

GitHub: orgs, repos, or full repo URLs

Wayback: domains or URLs to pull archived pages

DockerHub: orgs/users or full hub URLs

npm / PyPI: scopes or prefixes (@org, org-, org)

CI/CD: artifact or log URLs

APK / IPA: package or bundle IDs

Click Save Scope β†’ writes scope.yaml.

  1. Run a scan
    In the Engine Scan panel:

Select adapters (e.g., github, wayback, dockerhub)

Toggle Redact secrets βœ…

Adjust concurrency (default: 6)

Click Run scan + AI summary

Logs stream live; results are written into ./data/out/.

  1. Review results
    findings.jsonl β†’ machine-readable findings

findings.sarif.json β†’ import into security dashboards

summary.md β†’ AI-generated analysis for humans

Export everything with one click: Download Export Pack

⚑ Example Scope (Mozilla Bug Bounty)

github:
targets:
- "mozilla"
- "mozilla-mobile"
- "mozilla-services"
- "mozilla-releng"
- "mozilla-iot"

wayback:
targets:
- "github.com/mozilla"
- "hg.mozilla.org"
- "archive.mozilla.org"

dockerhub:
targets:
- "mozilla"

npm:
targets:
- "@mozilla"
- "mozilla-*"

pypi:
targets:
- "mozilla*"
πŸ“¦ Roadmap
Custom regex packs (user-supplied)

Multi-engine federation (run several adapters in parallel clusters)

Timeline mode for Wayback leaks

Export directly to HackerOne / Bugcrowd report templates

Preset scopes for popular programs (Mozilla, Google, etc.)

⚠️ Disclaimer
This tool is for educational and defensive research only.
Use it only on assets that are in-scope and authorized by a bug bounty program or your own organization.
Never target systems you don’t have permission to test.

Made with 🧑 by the GEARZ crew (solo dev, me and a cat 🐈).

README history 6 revisions

Every night we snapshot the README of every dataset in the catalog. When the SHA changes we archive the new version and diff it against the last. This is the evolving thought record of the alignment-data field: what the author decided to say about the corpus, and how that framing shifted over time.

  1. 2025-08-18docs(hub): add dataset card metadata87d9f093.4 KB
    Loading...
  2. 2025-08-18docs: keep local README; resolve HF stub mergeb46f3713.3 KB
    Loading...
  3. 2025-08-18initial commitf9f90a924 B
    Loading...
  4. 2025-08-18docs: update README with inline image renderc3e3a763.3 KB
    Loading...
  5. 2025-08-18feat: VHS theme background + UI restore + globals.css update + detailed README7a65dcd3.1 KB
    Loading...
  6. 2025-07-28Initial commit from Create Next Appde28e0e1.4 KB
    Loading...

Used by abliterated models none yet

No abliterated models in our catalog have declared this dataset in their YAML metadata yet. This can mean the dataset is used but not declared, is used outside abliteration workflows, or is new. The link will populate automatically as models are indexed.

Related datasets same stage Β· misc

Read further

Metadata

License
mit

Repository

Download the dataset

0

Copy any of these snippets into your notebook or terminal. All three fetch directly from Hugging Face using your own credentials.

Recommended: datasets library (Python)
from datasets import load_dataset
ds = load_dataset("Gearsoldier0731/gearz-leakfinder")
Raw snapshot (Python)
from huggingface_hub import snapshot_download
snapshot_download(repo_id="Gearsoldier0731/gearz-leakfinder", repo_type="dataset")
git clone (requires git-lfs)
git clone https://huggingface.co/datasets/Gearsoldier0731/gearz-leakfinder
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration