Discover. Confirm. Exploit. Report. Learn.
You give it a URL. It gives you the database.
And it gets smarter every time it runs.
Getting Started • How It Works • Features • Config • Learning DB • Hooks
SleepyWalker is a fully autonomous SQL injection pipeline that chains crawling, heuristic analysis, AI-powered confirmation, and sqlmap exploitation into a single command — and gets smarter with every scan through a persistent learning database.
sleepywalker -url https://target.internal -operator "you" -engagement-id "PT-2024-042"That's it. It crawls. It probes. It confirms. It dumps. It reports. It learns.
# Clone & build
git clone https://github.com/bidhata/SleepyWalker.git
cd SleepyWalker
go build -o sleepywalker ./cmd/
# Run your first scan (dry-run = no exploitation)
./sleepywalker -url https://target.example.com -dry-runRequirements: Go 1.23+ • sqlmap in PATH • (Optional) OpenRouter API key for AI mode
┌─────────────────────────────────────────┐
│ YOU RUN ONE COMMAND │
└────────────────────┬────────────────────┘
│
┌────────────────────▼────────────────────┐
│ Learning DB loaded from prior scans │
│ → enriches payloads & error signatures │
└────────────────────┬────────────────────┘
│
┌────────────────────▼────────────────────┐
│ PRE-SCAN: WAF Detection │
│ Fingerprints Cloudflare, AWS WAF, │
│ ModSecurity, Imperva, Akamai... │
└────────────────────┬────────────────────┘
│
┌──────────────────────────────▼──────────────────────────────┐
│ PHASE 1: Discovery │
│ │
│ 🕷️ Recursive HTML Crawl 🌐 JS-Rendered (chromedp) │
│ 📋 OpenAPI/Swagger Import 📡 Header/JSON/Body injection │
│ 🔗 Path segment injection 📎 Multipart forms │
│ │
│ → Built-in + learned payloads per injection context │
│ → Built-in + learned DB error signatures │
│ → Phase 1a: error-based 1b: boolean-blind 1c: time-based │
└──────────────────────────────┬──────────────────────────────┘
│
┌──────────────────────────────▼──────────────────────────────┐
│ PHASE 2: Confirmation │
│ │
│ 🤖 AI Mode (OpenRouter/gpt-4o-mini) │
│ → Structured JSON verdict with confidence score │
│ → Retry with exponential backoff │
│ │
│ 🔬 Offline Deep Validation (11 techniques) │
│ → Boolean-blind (Jaccard similarity) │
│ → Time-blind (multi-round statistical median) │
│ → Error-based, UNION-based, DB-specific probes │
│ → Content-length delta, status correlation │
│ → Second-order, Out-of-band DNS resolution │
│ │
│ Composite weighted confidence score → 0.0 – 1.0 │
└──────────────────────────────┬──────────────────────────────┘
│
┌───────────▼───────────┐
│ OPERATOR CONFIRMS │
│ [y/N] prompt │
└───────────┬───────────┘
│
┌───────────────────────────────▼─────────────────────────────┐
│ PHASE 3: Exploitation │
│ │
│ 🗡️ sqlmap --batch --dump │
│ → Auto-selects tamper scripts per WAF │
│ → Forwards cookies, headers, proxy │
│ → Configurable --risk and --level │
└───────────────────────────────┬─────────────────────────────┘
│
┌─────────────────────▼─────────────────────┐
│ 📊 REPORTING │
│ │
│ HTML • JSON • SARIF 2.1.0 │
│ CWE-89 • CVSS 9.8 • Remediation │
└─────────────────────┬─────────────────────┘
│
┌─────────────────────▼─────────────────────┐
│ 🧠 LEARNING DB UPDATED │
│ Confirmed payloads scored • FPs noted │
│ Host profiles • WAF fingerprints │
│ Next scan is faster and more accurate │
└───────────────────────────────────────────┘
| Feature | SleepyWalker | Basic Scanners |
|---|---|---|
| AI-powered false positive elimination | ✅ | ❌ |
| 11-technique deep validation (no AI needed) | ✅ | ❌ |
| 200+ advanced payloads (WAF bypass, polyglot, DB-specific) | ✅ | ❌ |
| 12 database engine detection (MySQL → MongoDB) | ✅ | ❌ |
| Self-learning pattern database | ✅ | ❌ |
| WAF-aware payload filtering + tamper auto-selection | ✅ | ❌ |
| Full audit trail (JSONL) for compliance | ✅ | ❌ |
| Scope control (regex + CIDR) | ✅ | ❌ |
| JS-rendered page crawling | ✅ | ❌ |
| OpenAPI/Swagger endpoint import | ✅ | ❌ |
| Multipart form + path segment injection | ✅ | ❌ |
| SARIF output for CI/CD integration | ✅ | ❌ |
| Plugin/hook system (5 lifecycle phases) | ✅ | ❌ |
| Interactive exploitation consent | ✅ | ❌ |
| Adaptive rate limiting with backoff | ✅ | ❌ |
| TOML config file support | ✅ | ❌ |
| MCP server for AI agent integration | ✅ | ❌ |
| Single binary, zero runtime deps | ✅ | ❌ |
| Location | Methods | Detection |
|---|---|---|
| Query parameters | GET, DELETE | Error, Boolean, Time, UNION, Conditional |
| Form body | POST, PUT, PATCH | Error, Boolean, Time, UNION, Conditional |
| JSON body | POST, PUT, PATCH | Error, Boolean, Time, UNION, Conditional |
| HTTP headers | User-Agent, Referer, X-Forwarded-For, Cookie | Error, Boolean, Time |
| Multipart fields | POST | Error, Boolean |
| URL path segments | GET (integers, UUIDs, hex IDs) | Error, Boolean |
| Engine | Error Signatures | Time Payloads | Specific Probes |
|---|---|---|---|
| MySQL / MariaDB | 11 | SLEEP, BENCHMARK | EXTRACTVALUE, UPDATEXML, EXP, JSON_KEYS |
| PostgreSQL | 9 | pg_sleep, generate_series | CAST, dblink_connect |
| Microsoft SQL Server | 10 | WAITFOR DELAY | CONVERT, HAVING, xp_dirtree |
| Oracle | 11 | DBMS_PIPE, DBMS_LOCK | UTL_INADDR, CTXSYS, UTL_HTTP, XMLType |
| SQLite | 8 | RANDOMBLOB (CPU) | sqlite_master, typeof |
| IBM DB2 | 6 | — | sysibm.sysdummy1 |
| SAP HANA | 4 | — | SYS.M_DATABASE, DUMMY |
| Firebird / InterBase | 4 | — | rdb$database |
| Informix | 4 | — | systables |
| Sybase / SAP ASE | 4 | — | sybsystemprocs |
| MongoDB (NoSQL) | 4 | — | $gt, $ne, $where, $regex |
| CockroachDB | 3 | — | version() pattern |
| H2 Database | 3 | — | H2VERSION() |
| Category | Count | Techniques |
|---|---|---|
| Syntax breakers | 9 | Quote types, escapes, parenthetical |
| Boolean injection | 15 | OR/AND in all quote contexts, auth bypass |
| Time-based blind | 10 | Per-DB sleep functions, BENCHMARK, CPU DoS |
| Error extraction | 10 | EXTRACTVALUE, CAST, EXP, GTID, POLYGON |
| UNION-based | 10 | NULL column enumeration, hex markers |
| WAF bypass (comments) | 7 | /*!50000*/, /**/, versioned comments |
| WAF bypass (encoding) | 8 | Case alternation, URL-encode, double-encode |
| WAF bypass (whitespace) | 4 | Null bytes, CRLF, tab, vertical tab |
| Stacked queries | 5 | Multi-statement, DECLARE, xp_cmdshell |
| Conditional error blind | 5 | CASE WHEN + divide-by-zero |
| ORDER BY / GROUP BY | 5 | Column count discovery, subquery injection |
| Polyglot / chained | 8 | Multi-context, XOR+sysdate, nested concat |
| Authentication bypass | 7 | admin'--, LIMIT/OFFSET, UNION credentials |
| Bit manipulation | 6 | AND/XOR/DIV/MOD bitwise operators |
| DNS/HTTP exfiltration | 3 | UNC path + VERSION(), UTL_HTTP |
| Privilege escalation | 4 | file_priv, sysadmin, super_priv checks |
| NoSQL / MongoDB | 6 | $gt, $regex, $where JS, $lookup pipeline |
| DB-specific probes | 20+ | DB2, Oracle, MSSQL, PostgreSQL, Firebird, H2, Informix, HANA |
Built for authorized red team operations with guardrails:
- Scope enforcement — regex and CIDR whitelist; blocks out-of-scope targets at startup
- Dry-run mode — full pipeline minus exploitation; perfect for CI/CD gates
- Exploitation consent — interactive
[y/N]prompt listing every target before Phase 3 - Request budget — hard cap on total HTTP requests (CAS-atomic) to prevent accidental DoS
- Adaptive throttling — auto-backs off on 429/503, exponential backoff with recovery
- WAF-aware rate limiting — detects blocked tokens, filters futile payloads, respects WAF limits
- Graceful shutdown — Ctrl+C flushes audit logs + saves learning DB, no data loss
- Full audit trail — every action logged as JSONL with operator identity
SleepyWalker maintains a persistent learning database at ~/.sleepywalker/learningdb.json that gets smarter with every scan.
| Category | What's Stored | Impact |
|---|---|---|
| Error signatures | New DB error patterns not in the built-in list | Detected on future scans automatically |
| Successful payloads | Payloads that confirmed SQLi, scored by success rate | Prioritised in future probes |
| WAF fingerprints | WAF header/body patterns seen in the wild | Faster WAF detection |
| Host profiles | Per-host: DB engine, WAF, injectable params | Skip known-clean endpoints on rescan |
| False positives | Host+param combos that fired but weren't injectable | Skipped after 5 false hits |
First scan: Built-in 10 payloads → 1 confirmed → DB records the winning payload
Second scan: 11 payloads (built-in + learned) → winning payload tried first
Tenth scan: Learned payloads with >80% hit rate sorted to front → faster detection
# Use a custom DB path (useful for team-shared DBs)
sleepywalker -url https://target.internal -ldb-path /shared/team-learningdb.json
# Default path
~/.sleepywalker/learningdb.json (Linux/macOS)
%APPDATA%\.sleepywalker\learningdb.json (Windows)sleepywalker -url https://target.internal \
-scope ".*\.internal$" \
-operator "jsmith" \
-engagement-id "PT-2024-042" \
-depth 3 \
-threads 8 \
-delay 200 \
-max-requests 5000 \
-output-format all \
-log-dir ./logssleepywalker -config engagement.toml[scope]
regex = ".*\\.internal$"
cidrs = ["10.0.0.0/8"]
[scan]
depth = 3
threads = 8
delay_ms = 200
max_requests = 5000
[auth]
cookie = "session=abc123"
headers = ["Authorization: Bearer <token>"]
[output]
format = "all"
[audit]
operator = "jsmith"
engagement_id = "PT-2024-042"
log_dir = "./logs"
[[hooks]]
name = "notify-slack"
phase = "post-report"
command = "python3"
args = "scripts/notify_slack.py"
timeout = 15Full example:
examples/engagement.toml
| Flag | Default | Description |
|---|---|---|
-url |
(required) | Target URL |
-config |
TOML config file path | |
-scope |
Regex scope restriction | |
-scope-cidr |
CIDR whitelist (repeatable) | |
-dry-run |
false |
Report without exploitation |
-max-requests |
0 |
Request budget (0 = unlimited) |
-depth |
0 |
Crawl depth (0 = unlimited) |
-threads |
4 |
Concurrency |
-delay |
0 |
Request delay (ms) |
-js-render |
false |
Headless browser crawl |
-swagger-url |
OpenAPI spec URL | |
-insecure |
false |
Skip TLS verification |
-cookie |
Cookie header | |
-header |
Custom header (repeatable) | |
-proxy |
HTTP/SOCKS5 proxy | |
-ai-provider |
openrouter |
AI: openrouter, bedrock, local |
-sqlmap-path |
(auto) | sqlmap binary path |
-risk |
2 |
sqlmap risk (1-3) |
-level |
3 |
sqlmap level (1-5) |
-output-format |
html |
html, json, sarif, all |
-operator |
Operator ID | |
-engagement-id |
Engagement reference | |
-log-dir |
Audit log directory | |
-hooks-dir |
Hook scripts directory | |
-ldb-path |
~/.sleepywalker/learningdb.json |
Learning DB path |
Extend SleepyWalker without touching source code. Hooks execute at 5 lifecycle phases:
pre-scan → post-discovery → post-confirm → post-exploit → post-report
hooks/
├── pre-scan_validate-engagement.sh
├── post-discovery_count-endpoints.py
├── post-report_notify-slack.py
└── post-report_upload-s3.sh
sleepywalker -url https://target.internal -hooks-dir ./hooks/Every hook receives full scan context as JSON on stdin:
{
"phase": "post-report",
"target_url": "https://target.internal",
"operator": "jsmith",
"engagement_id": "PT-2024-042",
"timestamp": "2024-01-15T10:30:00Z",
"data": {
"reports": ["./dumps/target_internal/report.html"],
"output_dir": "./dumps/target_internal"
}
}| Format | Use Case | Integration |
|---|---|---|
| HTML | Human-readable report with dark UI | Email, browser |
| JSON | Machine processing, dashboards | SIEM, custom tooling |
| SARIF 2.1.0 | CI/CD security gates | GitHub Code Scanning, Azure DevOps, Jira |
Output lands in ./dumps/<target-host>/:
dumps/target_internal/
├── report.html
├── report.json
├── report.sarif.json
└── <hash>/
└── *.sql, *.csv
SleepyWalker ships a Model Context Protocol server so AI agents (Claude, Amazon Q, etc.) can invoke scan capabilities as tools.
go build -o sleepywalker-mcp ./mcp/Register in your MCP client config:
{
"mcpServers": {
"sleepywalker": {
"command": "/path/to/sleepywalker-mcp"
}
}
}Tools exposed: sleepywalker_scan • sleepywalker_discover • sleepywalker_waf_detect • sleepywalker_validate • sleepywalker_report
SleepyWalker/
├── cmd/main.go # CLI + pipeline orchestration
├── mcp/main.go # MCP server for AI agent integration
├── internal/
│ ├── config/
│ │ ├── config.go # Runtime config, scope validation, TLS
│ │ └── configfile.go # TOML config loader + CLI merge
│ ├── hooks/hooks.go # Plugin/hook lifecycle system
│ ├── learningdb/db.go # Persistent learning database
│ ├── scanner/
│ │ ├── scanner.go # Crawl engine (HTML + recursive)
│ │ ├── jscrawl.go # Headless browser (chromedp)
│ │ ├── swagger.go # OpenAPI/Swagger parser
│ │ ├── heuristic.go # Phase 1: payload injection + learning
│ │ ├── deepvalidate.go # Phase 2: 10-technique validation
│ │ └── waf.go # WAF fingerprinting
│ ├── ai/ai.go # Phase 2: AI analysis (multi-provider)
│ ├── sqlmap/sqlmap.go # Phase 3: sqlmap wrapper
│ ├── reporter/
│ │ ├── reporter.go # HTML report
│ │ └── jsonreport.go # JSON + SARIF
│ └── utils/
│ ├── audit.go # JSONL audit trail
│ ├── ratelimit.go # Adaptive rate limiter
│ ├── logger.go # Logging
│ └── fileutils.go # Output dirs
├── examples/
│ ├── engagement.toml # Sample config
│ └── post-report_notify-slack.py # Sample hook
├── .gitignore
├── go.mod
└── README.md
# Internal pentest with full audit trail
sleepywalker -url https://app.corp.internal \
-config pentest.toml \
-operator "krish" \
-engagement-id "RT-2024-087" \
-log-dir ./audit-logs
# CI/CD gate (dry-run + SARIF for GitHub)
sleepywalker -url https://staging.myapp.com \
-dry-run \
-output-format sarif \
-max-requests 1000
# Aggressive internal scan through proxy
sleepywalker -url https://legacy-app.internal \
-proxy socks5://127.0.0.1:9050 \
-risk 3 -level 5 \
-depth 5 -threads 16 \
-insecure
# JavaScript-heavy SPA with API spec
sleepywalker -url https://spa.internal \
-js-render \
-swagger-url https://spa.internal/api/docs/openapi.json \
-output-format all
# Team-shared learning DB
sleepywalker -url https://target.internal \
-ldb-path /team-share/sleepywalker-learningdb.json- Fork it
- Create your feature branch (
git checkout -b feat/amazing-feature) - Commit your changes (
git commit -m 'Add amazing feature') - Push to the branch (
git push origin feat/amazing-feature) - Open a Pull Request
Areas where help is appreciated:
- Additional WAF bypass techniques
- New deep validation heuristics
- Integration hooks for more platforms
- Learning DB export/import tooling
This tool is intended EXCLUSIVELY for authorized penetration testing.
Unauthorized use against systems you do not own or have explicit written permission to test is illegal and unethical. Always obtain proper authorization before scanning any target. The author accepts no liability for misuse.
By using this tool, you agree that you have legal authorization to test the target systems.
Built with obsessive attention to detail by Krish Paul
If this tool saved your red team hours of manual testing, consider giving it a ⭐
Go • OpenRouter AI • sqlmap • chromedp • MCP