SAST Scanner · Supply Chain Shield · Network Auditor · Build Intelligence · Skills System · MCP Server
Sentinel is a deterministic, local-first security intelligence platform. It scans source code for vulnerabilities, audits software supply chains, monitors network behavior for exfiltration attempts, observes build processes, and verifies system integrity — all without sending data to third-party services.
It also integrates with AI coding agents (Claude, Cursor, Cline, Windsurf, OpenCode, Roo, Gemini, Codex) via platform-specific skills and a Model Context Protocol (MCP) server, but Sentinel works standalone as a CLI security tool for any development workflow.
| I want to... | Go to |
|---|---|
| Install and try it | Quick Start |
| Understand what it does | Core Capabilities |
| See a real output | Demo |
| Understand the architecture | Architecture |
| Read the design principles | Design Philosophy |
| Compare with other tools | How Sentinel Compares |
| Find a use case | Use Cases |
| Learn the CLI commands | CLI Commands |
| See benchmark numbers | Benchmarks |
| Understand maturity level | Current Maturity |
| Read the full documentation | Documentation |
| Contribute | Contributing |
| Report a vulnerability | Security |
| See the roadmap | Roadmap |
| Area | Status |
|---|---|
| SAST (30 rules) | Production-ready |
| Supply chain audit | Production-ready |
| Network auditor | Production-ready |
| GitHub integration | Production-ready |
| Build intelligence | Production-ready |
| Red Team emulation | Research-grade (simulated) |
| Replay & regression | Implemented, needs external validation |
| Atomic RT validation | Pending (requires isolated VM) |
| CALDERA campaigns | Pending |
| External benchmarking | Pending |
1,040 automated regression and unit tests covering implemented functionality. CI runs on Node 20 + 22.
npm install -g sntnl-clinpx sntnl-cli scan ./srcgit clone https://github.com/javier20dev25/sentinel-cli.git
cd sentinel-cli
npm install
npm run build
node dist/cli/main.js scan ./src# Scan a project for vulnerabilities
sntnl scan ./src
# Audit all dependencies
sntnl audit-deps
# Start network monitoring
sntnl network start
# Observe a build
sntnl build observe "npm run build"
# Install skills for AI coding agents (optional)
sntnl install-skills30 deterministic rules covering: secrets detection, eval() usage, network access, environment variable exposure, command injection, SQL injection, prototype pollution, crypto misuse, obfuscation, and workflow poisoning.
sntnl scan <path>Multi-layer package auditing without installing anything:
| Layer | Mechanism |
|---|---|
| CVE lookup | OSV.dev batch query (up to 1,000 packages) |
| Typosquat detection | Damerau-Levenshtein + homoglyph detection |
| Provenance verification | SLSA attestation via npm attestation verify |
| Registry reputation | 6-factor scoring (age, maintainers, versions, deprecation, description, homepage) |
| Malicious patterns | Embedded secrets, suspicious install scripts |
sntnl audit-deps # Full dependency audit
sntnl verify-pkg <name> # Single package check
sntnl sbom # CycloneDX v1.5 SBOM generation
sntnl deps-tree <path> # Transitive dependency scanMonitors process execution, Git activity, network connections, and DNS queries to detect repository exfiltration patterns. Unlike traditional tools that inspect packet contents, Sentinel reconstructs behavioral chains (preparation → collection → packaging → exfiltration) to detect the shape of an attack regardless of the tools used.
31 behavior classifiers mapped to MITRE ATT&CK, with SHA-256 evidence chaining and timeline reconstruction.
sntnl network start # Start a live audit session
sntnl network stop # Stop and produce verdict
sntnl network status # Current session status
sntnl network history -l N # Last N sessions
sntnl network session <id> # Session detail
sntnl network export <id> # Export (json|markdown)Captures a complete forensic record of any build command: process tree (filtered to build descendants), file artifacts created during the build, and a SHA-256 integrity chain. Outputs a CLEAN/REVIEW verdict based on anomalous processes.
sntnl build observe "npm run build"
sntnl build observe "make -j4" --verbose
sntnl build observe "go build ./..." --json
sntnl build explain # Why the score is what it is
sntnl inspect # Evidence graph, centrality, dominatorssntnl pr-audit --repo R --pr N # Audit a single PR
sntnl workflow full-audit --repo R # Audit ALL PRs in one repoPR Bot auto-analyzes all open PRs across your repos, posts findings as PR comments with Check Run status.
| Command | Description |
|---|---|
sntnl integrity |
Chain-of-trust host verification (hash, PATH, vault, clock, manifest) |
sntnl doctor |
Dependency health and system diagnostics |
sntnl baseline create|diff |
System snapshots and drift detection |
sntnl permissions <pkg> |
Capability audit of installed packages |
OS-level package manager interception and git hooks:
sntnl guard enable # Intercept npm/yarn/pip installs
sntnl guard status # Check guard status
sntnl precommit install # Block commits with threats
sntnl prepush install # Block pushes with threats
sntnl install npm <pkg> # Scan then installsntnl memory --status # View vault status
sntnl memory --ingest <file> # Ingest cloud report
sntnl memory --findings # Query past findings
sntnl memory --threats # Threat correlationsClassifies and risk-assesses any token string — GitHub PATs, AWS keys, Stripe secrets, Slack tokens, and more. Produces risk scores, scope analysis, and actionable recommendations.
sntnl token-inspect <token> # Full classification + risk assessmentExport scan findings and reports in multiple formats for integration with external tools, CI pipelines, and compliance systems.
sntnl export json # Machine-readable JSON
sntnl export markdown # Human-readable Markdown
sntnl export pdf # Printable PDF report
sntnl export sarif # SARIF for GitHub Code Scanning
sntnl export policy # Policy configuration exportsntnl check-classified <path> # Pre-commit classified data check
sntnl env-encrypt <file> # AES-256-CBC encryption for .env files
sntnl env-decrypt <file> # Decrypt .env.enc back to plaintextConfigure security policies and track behavioral drift of package capabilities across versions.
sntnl policy # ci-mode, fail-closed, quarantine settings
sntnl drift # Track capability drift across versions
sntnl baseline create|diff # System snapshots and drift detectionAdvanced analysis of build process graphs: centrality measures, dominator trees, Bayesian inference shifts, and critical path detection.
sntnl graph history # Snapshot history with chain count trend
sntnl graph diff # Diff between latest two snapshots
sntnl graph analytics # Centrality, dominators, Bayesian, critical path
sntnl inspect # Evidence graph investigationExtended network auditor capabilities for advanced threat analysis:
sntnl network trusted # Manage trusted agent allowlist
sntnl network doctor # Health check, coverage, sensor drift
sntnl network blindspots # Record and manage detection failures
sntnl network campaign # Run validation campaigns
sntnl network benchmark # Benchmark history across engine versions
sntnl network replay # Replay sessions through detection pipeline
sntnl network record # Record real OS session
sntnl network corpus # Inspect corpus coveragesntnl redteam --list # 26 attack scenarios, 10 campaigns
sntnl redteam --coverage # Coverage matrix
sntnl atomic --list # 30+ Atomic RT tests mapped
sntnl atomic --dry-run # Preview without executing
sntnl coverage # MITRE ATT&CK matrix
sntnl replay list # Replay datasets
sntnl regression list # Regression suitessntnl hub launches a bilingual (English/Spanish) operations menu:
0. PR Bot — Auto-Analyze All Open PRs
1. Select Workspaces & Run Audits
2. System Doctor (Health Check)
3. Integrity Check
4. Permissions Audit
5. Scan Directory/File
6. Sentinel Guard & Configuration
7. Classified Documents
8. Manage Signal Vault (Memory)
9. Security — Integrity & Trust Policy
10. Network Auditor
11. Exit
- Replace EDR — Sentinel observes builds and monitors networks, not runtime
- Replace SAST tools — it complements them with build-time context
- Replace DAST — no dynamic testing of running applications
- Replace SCA — uses OSV for CVE lookup, not a full SCA replacement
- Auto-remediate — it reports and blocks (via guard/precommit), but does not fix code
- Detect zero-days — relies on observable patterns, not vulnerability research
- Analyze firmware/hypervisor — out of scope
Platform-specific instruction files that teach AI agents to call Sentinel for all security-relevant tasks. Available for 8 agent platforms:
sntnl install-skills # auto-detect agents
sntnl install-skills --list # show detected agents
sntnl install-skills --all # install for all platformsEach skill teaches the agent:
- Sentinel primacy — call
sntnl scanbefore reading code with the model - Evidence attachment — every security claim must include verbatim Sentinel output
- Trust hierarchy — Sentinel evidence (Tier 1) overrides model reasoning (Tier 4)
- Workflows — PR review, package audit, host integrity, full audit pipelines
| Agent | Format | File |
|---|---|---|
| Claude Code | Markdown | skills/adapters/claude/CLAUD.md |
| Cursor | YAML + MDC | skills/adapters/cursor/sentinel.mdc |
| Cline | Markdown | skills/adapters/cline/CLINE.md |
| Windsurf | Flat rules | skills/adapters/windsurf/.windsurfrules |
| OpenCode | Markdown | skills/adapters/opencode/SKILL.md |
| Roo Code | Markdown | skills/adapters/roo/ROO.md |
| Gemini CLI | YAML + triple-file | skills/adapters/gemini/GEMINI.md |
| OpenAI Codex | Markdown | skills/adapters/codex/CODEX.md |
Model Context Protocol server (sntnl mcp) exposing 17 tools as standard MCP tool calls:
| Category | Tools |
|---|---|
| SAST + Analysis | scan, verify-pkg, doctor, integrity, audit-deps, deps-tree, sbom |
| Security Gate | check-classified, trust-cache, policy |
| Intelligence | memory, threat-query, threat-correlate |
| GitHub PR Tools | gh-pr-list, gh-pr-view, gh-pr-diff, gh-repo-list |
sntnl mcp # stdio mode (default)
sntnl mcp --http --port 3003 # HTTP/SSE mode$ sntnl build observe "npm run build"
══════════════════════════════════════════════
SENTINEL BUILD OBSERVATION
══════════════════════════════════════════════
CLEAN 90/100 9.0s
What happened
──────────────
Tools: none detected
Processes: 0 observed
Build Identity
──────────────
Hermetic: 95/100
Reproducible: 0/100
Confidence: 0
Why this score
──────────────
-5 [MEDIUM] 235 named pipe(s) detected (inter-process communication not tracked)
-15 No build tools detected (may not be a build)
+5 Hermetic build (95/100)
+5 No network activity during build
Nothing requires immediate action.
$ sntnl pr-audit --repo javier20dev25/sntnl --pr 42
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✔ CLEAN │ PR #42 │ javier20dev25/sentinel-cli
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Title: feat: add new detection rule
Author: contributor123
Files: 3 changed
Lines: +45/-12
Rules: 30 SAST rules (code + secrets + filenames)
Status: CLEAN BILL OF HEALTH
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Sentinel exists because of six convictions:
- Evidence over signatures. Observe what happened, not what matches a pattern.
- Explainability over opaque scoring. Every verdict traces to observed evidence.
- Deterministic before probabilistic. 30 SAST rules before Bayesian inference.
- Replay before repetition. Capture once, validate many times.
- Observable systems produce better security. If you can't see it, you can't secure it.
- Offensive thinking improves defensive validation. Red Team validates Blue Team.
| Tool | What it does | Sentinel's approach |
|---|---|---|
| CodeQL | Static analysis (source code patterns) | Observes build behavior, not just code patterns |
| Snyk | Dependency vulnerability scanning | Multi-layer supply chain + behavioral analysis |
| Trivy | Container image scanning | Build-time observation, not image scanning |
| Falco | Runtime security (syscall monitoring) | Build-scoped observation with forensic detail |
| Sysmon | System telemetry collection | Reconstructs causal chains from telemetry |
| Sigstore | Artifact signing/verification | Captures provenance during build, verifies post-build |
Sentinel is not better than these tools. It does something different: it observes builds, monitors networks, and explains trust. It complements SAST, SCA, and EDR — it doesn't replace them.
| When | How |
|---|---|
| Pull Request review | sntnl pr-audit posts findings as PR comment |
| Release pipeline | sntnl build observe as quality gate |
| Incident investigation | sntnl build explain for causal analysis |
| Supply chain audit | sntnl audit-deps for dependency trust |
| Red Team validation | sntnl redteam for attack simulation |
| Regression detection | sntnl regression for automated validation |
| AI agent integration | sntnl mcp for 17 tool calls via MCP protocol |
| System integrity | sntnl integrity for chain-of-trust verification |
| Package interception | sntnl guard enable for OS-level npm/pip blocking |
| Threat history | sntnl memory for local threat intelligence vault |
| Token risk assessment | sntnl token-inspect to classify any token |
| Data protection | sntnl env-encrypt for .env file encryption |
| Graph analysis | sntnl graph analytics for build process insights |
| Network forensics | sntnl network record + replay for session analysis |
| Policy enforcement | sntnl policy for ci-mode, fail-closed, quarantine |
| SARIF integration | sntnl export sarif for GitHub Code Scanning |
sntnl scan <path> # SAST findings by severity
sntnl scan --staged # Scan staged git changes
sntnl audit-deps # Dependency audit (lockfile + OSV)
sntnl verify-pkg <name> # Single package check
sntnl sbom # CycloneDX v1.5 SBOM
sntnl deps-tree <path> # Transitive dependency scansntnl build observe <cmd> # Verdict + trust + explanation
sntnl build explain # Why the score is what it is
sntnl build mark-release # Mark current build as release baseline
sntnl build release # Show current release baseline info
sntnl build trust # Trust calibration: corpus stats, model state
sntnl build learning # Continuous learning pipeline, model versionssntnl graph history # Graph snapshot history with chain count trend
sntnl graph diff # Diff between latest two graph snapshots
sntnl graph analytics # Centrality, dominators, Bayesian shifts, critical path
sntnl inspect # Evidence graph, centrality, dominators
sntnl top # Top findings from recent buildssntnl network start # Start live monitoring
sntnl network stop # Stop and get verdict
sntnl network status # Current session status
sntnl network history -l N # Last N sessions
sntnl network session <id> # Session detail
sntnl network export <id> # Export session (json|markdown)
sntnl network trusted # Manage trusted agents
sntnl network doctor # Health check, coverage, sensor drift
sntnl network blindspots # Record detection failures
sntnl network campaign # Run validation campaigns
sntnl network benchmark # Benchmark history across engine versions
sntnl network replay # Replay recorded sessions
sntnl network record # Record real OS session for replay
sntnl network corpus # Inspect corpus coverage vs canonical profiles
sntnl redteam --list # 26 attacks, 10 campaigns
sntnl atomic --list # 30+ Atomic RT tests mapped
sntnl coverage # MITRE ATT&CK matrixsntnl pr-audit --repo R --pr N # Audit a single PR
sntnl workflow full-audit # Audit ALL PRs in a repo
sntnl workflow pr-review # Audit + post results as PR commentsntnl guard enable # Intercept npm/yarn/pip
sntnl guard status # Check guard status
sntnl precommit install # Block commits with threats
sntnl prepush install # Block pushes with threats
sntnl install npm <pkg> # Scan then installsntnl check-classified <path> # Verify files for classified data
sntnl env-encrypt <file> # Encrypt .env with AES-256-CBC
sntnl env-decrypt <file> # Decrypt .env.enc back to plaintextsntnl token-inspect <token> # Classify and risk-assess any token
# (GitHub PAT, AWS key, Stripe, Slack, etc.)sntnl export json # Export findings as JSON
sntnl export markdown # Export findings as Markdown
sntnl export pdf # Export findings as PDF
sntnl export sarif # Export findings as SARIF
sntnl export policy # Export policy configurationsntnl replay list # Replay datasets
sntnl regression list # Regression suites
sntnl baseline-pro list # Baseline profiles
sntnl stress config # Stress testingsntnl policy # Configure security policies
# (ci-mode, fail-closed, quarantine)
sntnl drift # Track behavioral drift across versions
sntnl baseline create|diff # System snapshots and drift detectionsntnl mcp # MCP server (17 tools)
sntnl hub # Interactive operations menu
sntnl install-skills # Install for 8 AI agents
sntnl guide # Interactive user guide
sntnl policies # Show security policy & guidelinessntnl integrity # Chain-of-trust verification
sntnl doctor # System health check
sntnl permissions <pkg> # Capability audit
sntnl memory --status # Threat vault statussntnl benchmark # SAST precision/recall
sntnl network benchmark history # Detection pipeline metrics| Layer | Scenarios | Pass Rate |
|---|---|---|
| Calibrated corpus | 39 | 79.5% |
| Blind corpus #1 | 15 | 60.0% |
| Blind corpus #2 | 14 | 92.9% |
| Blind corpus #3 | 14 | 85.7% |
| Replay corpus | 200 | 80.0% accuracy, 100% recall |
| Tier | Source | Confidence |
|---|---|---|
| Tier 1 | Sentinel telemetry (ETW, eBPF, auditd) | High |
| Tier 2 | Sentinel inference (graph, Bayesian, trust) | Medium |
| Tier 3 | External feeds (OSV, MITRE, registry) | Variable |
| Tier 4 | Model reasoning (when used via MCP/skills) | Lowest |
Sentinel evidence (Tier 1) overrides model reasoning (Tier 4).
| Document | Description |
|---|---|
| Architecture | 938-line technical architecture |
| Build Intelligence | Complete feature reference |
| Build Flight Recorder | Build observation system |
| Network Auditor | Behavior-based detection |
| Behavior Engine | 31 behavior classifiers |
| Risk Engine | Risk scoring methodology |
| Trust Model | Evidence hierarchy and calibration |
| Skills | AI agent integration guide |
| CI/CD Security | CI pipeline security |
| Corpus | Canonical profile system |
| Replay System | Session replay architecture |
| Recording Guide | How to record sessions |
| Ground Truth | Validation methodology |
| Limitations | Known limitations |
| Classification Policy | Detection policy contract |
| Technical Report | 801-line technical report |
| Legal | Legal notices and attributions |
| Contributing | Development guide |
| Security | Vulnerability reporting |
| Roadmap | Direction without dates |
- Node.js >= 20.0.0
- npm >= 9
- gh CLI (for GitHub PR tools — optional)
Business Source License 1.1 — see LICENSE.
Changes to GPL v2.0 on 2030-05-20.
Built for developers who take security seriously.