Lictor-AI

skill
Security Audit
Fail
Health Warn
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Fail
  • Hardcoded secret — Potential hardcoded credential in .github/workflows/ci.yml
  • rm -rf — Recursive force deletion command in airlock/package.json
  • rm -rf — Recursive force deletion command in airlock/src/audit/english.ts
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Open-source AI security suite — Shield (Chrome ext) + Sentinel (npm/pip SDK) + Guardian (hosted dashboard) + Claude Code skills. Apache 2.0.

README.md

Lictor AI

The bodyguard your AI didn't ship with.

CI
License: Apache 2.0
npm version
PyPI version
Latest scanner release

🆕 patrol-scanner v0.2 just shipped — false-positive elimination across 5 providers (CloudFront, Azure CloudApp, GitHub Pages, Netlify, Heroku) with new header_fp_blocklist / status_required / verifier mechanism. 16 regression tests, 3 production FPs live-verified as filtered. Release notes →

Watch the Lictor demo (3 min)

Lictor is an open-source AI security suite, organized as three pillars on one shared engine — easy to hold in your head:

🤖 Lictor for AI — secure the AI you use

Drops into every AI client you already use — Claude Code, Cursor, Codex, Antigravity, Windsurf, and the rest of the top-10 — and guards each boundary an AI agent can cross:

  • Shield — Chrome extension: guards the AI-built app in your browser
  • Sentinel — SDK (npm/pip): guards the prompt (injection + PII)
  • Airlock — action broker: guards the action (MCP tool-calls + shell; observe → enforce)
  • Skills · VS Code / Cursor ext · Studio — guard your dev loop (/lictor-security-check and friends, in every client)

🏢 Lictor for Business — secure the company · 2026–27 roadmap

The security an SMB needs but can't staff a team for:

  • Domain Guard — identities & Active Directory: stale passwords, over-privileged accounts, open shares
  • Isolation — the red/black sandbox: a pre-isolated machine to run & test AI agents off your real network, no env-build required
  • Audit & Rotate — compliance evidence + guided key rotation (built on Guardian + the lictor-rotate skill)

🛰️ Lictor Patrol — scan the world

  • v3 scanner + Patrol — 38-module external attack-surface recon; the engine behind "we scanned 30,000 vibe-coded apps." Apache 2.0, runs locally.

Under all three sits the platform: Lictor Core (a Rust crate compiling to native + WASM, plus the AUDIT.json output standard), the lictor CLI, and Lictor Guardian (the dashboard every product reports into — one audit trail, one place to prove you're secure). One engine, so every new check ships everywhere at once.

May 2026 update — 276 disclosures sent, 13 false-positive classes catalogued

In the four weeks before public launch, Lictor's scanners produced:

  • 276 verified disclosures to affected organisations (184 via GitHub public process + 92 via direct-email CVD)
  • 7,736 hosts scanned across 12 countries / 5 industries (finance, healthcare, government, crypto, AI infrastructure)
  • 38 scanner modules with anti-FP rigor (HEAD-only canaries, vendor-fingerprint detection, content-quality validation, baseline-compare classifiers)
  • 13 false-positive classes discovered and filtered (see v3/docs/fp-catalog.md)
  • 0 victim names published publicly — all disclosures routed privately under 60-day Coordinated Vulnerability Disclosure norms

The validated finding count after anti-FP filtering: 44 customer-app sourcemap exposures with revealed source code (down from 292 raw initial hits), 33 spoofable enterprise email domains, 11 publicly-listable cloud buckets, 14 EOL software stacks, 4 RDP/FTP/SMB services exposed direct-to-internet.

Full transparency at lictorai.com/transparency.

Status

🤖 Pillar 1 — Lictor for AI (shipped; the Oct 6 launch)

Component What it is Status
shield/ Chrome MV3 extension ✅ WASM-backed audits, real popup verdict, loads unpacked
sentinel/ TypeScript SDK (npm: @lictor/sentinel) ✅ 159 tests, real wrap(), 3 checks (prompt-injection / pii-leak / secrets-in-input)
sentinel-py/ Python SDK (PyPI: lictor-sentinel) ✅ 11 tests, Python parity, zero runtime deps
airlock/ AI action broker (npm: @lictor/airlock) ✅ 23 tests, observe→enforce, MCP + shell, 5 policy rule families, Guardian telemetry
skills/ Claude Code plugin suite ✅ 4 plugins, one-line install.sh, portable to Cursor/Codex
vscode-extension/ VS Code / Cursor extension ✅ Packages to .vsix; Marketplace + Open VSX at launch
studio/ Tauri desktop app 🚧 Builds local .dmg/.app; signed release Q1 2027

🏢 Pillar 2 — Lictor for Business (2026–27 roadmap)

Component What it is Status
domain-guard/ Identity & Active Directory posture 🗺️ Spec'd — stale passwords, over-privileged accounts, open shares
isolation/ Red/black AI sandbox machine 🗺️ Spec'd — run/test AI agents off the real network
Audit & Rotate Compliance evidence + key rotation ✅ Foundations exist (Guardian audit export + lictor-rotate skill); productizing

🛰️ Pillar 3 — Lictor Patrol (scanner shipped; autonomous Patrol building Q3)

Component What it is Status
v3/ External attack-surface scanner battery ✅ 38 scanner modules, anti-FP rigor (13 FP classes catalogued), 276 disclosures sent
disclosures/ Coordinated Vulnerability Disclosure pipeline ✅ Organised by tier (queue / courtesy-sourcemaps / bounty-eligible / verified-fp)

⚙️ The platform (shared by all three)

Component What it is Status
core/ Rust check engine, native + WASM + AUDIT.json standard ✅ 45 tests, native + WASM builds, Apache 2.0
cli/ lictor terminal binary lictor audit . --format human/json/markdown + --fail-on for CI
guardian/ Hosted dashboard — the one audit trail every product reports into ✅ 12 integration tests, magic-link auth, audit-log export, Slack webhook, docker compose up
landing/ Marketing site (lictorai.com) ✅ Multi-page: hero, /business, /in-the-wild, /transparency, /scan, /patterns, /compliance

Public launch target: Tuesday October 6, 2026.

Download & install

Every component installs today. SDKs/extensions publish to their public registries at the Oct 6, 2026 launch — until then, install from source or from the artifacts below. Build all downloadable artifacts at once with make package (→ dist/release/).

Component Install today At launch (registry)
CLI (lictor) curl -fsSL https://raw.githubusercontent.com/Raffa-jarrl/Lictor-AI/main/scripts/install.sh | bash — grabs the release binary, or builds from source cargo install lictor-cli / brew install lictor
Skills (Claude Code) git clone https://github.com/Raffa-jarrl/Lictor-AI && bash Lictor-AI/skills/install.sh — (always source)
Sentinel (npm) cd sentinel && pnpm install && pnpm build && npm packnpm i ./lictor-sentinel-*.tgz npm install @lictor/sentinel
Sentinel (PyPI) pip install ./sentinel-py (PEP 517 source install) pip install lictor-sentinel
Airlock (npm) cd airlock && npm install && npm run buildnode dist/cli.js --help npm i -g @lictor/airlock / npx @lictor/airlock
Shield (Chrome) Download lictor-shield-*.zip from Releases → unzip → chrome://extensions → Load unpacked Chrome Web Store
VS Code / Cursor Download lictor-*.vsix from Releasescode --install-extension lictor-*.vsix Marketplace + Open VSX
Guardian (self-host eval) cd guardian && docker compose uphttp://localhost:3100 hosted at lictorai.com
Studio (desktop) make studio-build.dmg/.app (unsigned; right-click→Open) signed .dmg Q1 2027

make install installs the CLI + Claude Code skills locally in one step. make package builds every artifact above into dist/release/. Guardian's source is source-available (run-it-yourself for dev/eval; not for hosting-as-a-service).

Real-world impact

Lictor's scanners are designed to find real security issues in real codebases. Open-source maintainers across these projects have received Lictor-filed reports — many have closed them with fixes or explicit acknowledgements:

"thanks for flagging the trigger pattern. The above workflow was obsolete and due to be removed, and done by PR #126"
aws-neuron/aws-neuron-samples (Amazon AWS Neuron team)

Other projects that have actioned reports from Lictor's automated patrols:

All reports filed via Lictor's open-source scanners (scripts/patrol-*.py, Apache 2.0). The same patterns are running against your code if you /lictor-security-check your project today.

Try the Claude Code plugins (60 seconds)

If you have Claude Code installed:

git clone https://github.com/Raffa-jarrl/Lictor-AI ~/Code/lictor
mkdir -p ~/.claude/skills
cp -r ~/Code/lictor/skills/lictor-* ~/.claude/skills/

Then in any Claude Code session, from your project's root directory:

/lictor-security-check

Claude reads your project, runs 7 checks, writes a SECURITY-AUDIT.md, and tells you in plain English what to fix.

Try the native CLI

git clone https://github.com/Raffa-jarrl/Lictor-AI
cd lictor

# One-time setup
brew install rustup wasm-pack    # macOS
rustup default stable
rustup target add wasm32-unknown-unknown

# Audit any URL
cargo run --release --example audit -- https://your-app.com -o report.md

A self-contained vulnerable demo lives in examples/vulnerable-demo/ — spin it up with python3 -m http.server 8765 and audit it to verify the engine works on a known-bad target.

Try the Chrome extension

cd shield
pnpm install
pnpm wasm     # builds lictor-core to WASM
pnpm build    # produces shield/dist/

In Chrome → chrome://extensions → Developer Mode → Load unpacked → select shield/dist/.

Visit any AI-built site (Vercel/Supabase/OpenAI/Anthropic fingerprints) and the toolbar badge reports findings.

Why

The AI security market is dominated by enterprise-only point solutions. Every one of them assumes you can wait for procurement, sign a 6-figure contract, and integrate over 3 months. None of them does anything for the consumer staring at a vibe-coded SaaS asking "is this actually safe?"

Lictor flanks them. Free for individuals. Free for OSS. Paid for teams. Enterprise where the money is. One engine across all of them so every new check ships everywhere at once.

In a sentence: safety infrastructure for the AI agent era.

License

Apache 2.0 for the whole suite — core/, cli/, shield/, sentinel/, sentinel-py/, airlock/, skills/, vscode-extension/, studio/, v3/, and the Business-pillar domain-guard/ + isolation/. The one exception: Guardian — its hosted service code is published for transparency but licensed for hosted use through lictorai.com only (see guardian/LICENSE).

See LICENSE + NOTICE.

Security

If you've found a vulnerability in Lictor itself, don't open a public issue — see SECURITY.md. False-negative reports (real bugs Lictor missed in your code) are the most valuable kind of report we get — use the false-negative template.

Contributing

Not yet broadly open. See CONTRIBUTING.md for what we DO want right now (bug reports, false-negative reports, pattern requests).

CODE_OF_CONDUCT.md applies.

What's in this repo

.
│  ── 🤖 Pillar 1: Lictor for AI ──
├── shield/           — Chrome extension (guards the app)
├── sentinel/         — TypeScript SDK (guards the prompt)
├── sentinel-py/      — Python SDK
├── airlock/          — AI action broker (guards the action: MCP + shell)
├── skills/           — Claude Code plugin suite (4 skills) + Cursor adapter
├── vscode-extension/ — VS Code / Cursor extension
├── studio/           — Tauri desktop app
│  ── 🏢 Pillar 2: Lictor for Business (roadmap) ──
├── domain-guard/     — identity & Active Directory posture
├── isolation/        — red/black AI sandbox machine
│  ── 🛰️ Pillar 3: Lictor Patrol ──
├── v3/               — external attack-surface scanner battery
├── disclosures/      — coordinated vulnerability disclosure pipeline
│  ── ⚙️ Shared platform ──
├── core/             — Rust crate, the shared engine + AUDIT.json
├── cli/              — `lictor` terminal binary
├── guardian/         — Next.js + Postgres dashboard (the one audit trail)
├── landing/          — Marketing site (HTML, deploys to lictorai.com)
├── brand/            — Logo SVG + rendered PNG icon variants
├── docs/             — Specs (sentinel-api, guardian-schema, wire-format) + launch/
└── examples/         — Vulnerable-by-design fixtures

Built by

A 20-year cybersecurity engineer + Claude. Solo build, building in public.

If this saved you from a bad day: star the repo. That's how solo open-source projects get measured.

Reviews (0)

No results found