See what's inside a repo before you run it.
MAYA Sentinel is a local repository ZIP scanner for bounded static analysis. It inspects archive structure and source signals, then produces public-safe Markdown, HTML, and JSON receipts. It never executes repository code, never installs dependencies, and never phones home.
Sentinel cannot prove that a repository is safe. It identifies static signals that help a human decide what deserves deeper review.
Upload a ZIP and read the decision brief. Every scan runs locally and stays local.
Pick a focus before you drop the ZIP. Every mode runs the same local static scan; the report just leads with what matters for your question.
| Mode | Question it answers |
|---|---|
| Safety Scan | Is this safe to run? Credentials, binaries, install hooks, filesystem and process risk. |
| Phishing / Impersonation | Is this a scam? Shortlinks, lookalike domains, scam tracking stacks, credential-harvest and wallet-drainer wording. |
| Dependency & Supply Chain | What does this pull in? Dependencies, lockfiles, install hooks, version drift. |
| AI / Agent Surface | Does this drive agents? MCP servers, agent instructions, prompts, workflows. |
| Exfil & Telemetry | Does this phone home? Webhook endpoints, tracking stacks, data collection. |
| Archive Safety | Is the ZIP itself hostile? Traversal, bombs, symlinks, path tricks. |
CLI scans take the same focus flag:
python maya_lens_server.py --scan path/to/repository.zip --mode phishing- ZIP traversal, path collision, symlink, device-name, compression-ratio, and extraction-budget hazards, the archive tricks that hide malware in plain sight
- Install hooks and dependency manifests, what runs when you install
- Credential-shaped values, with redaction, not exposure
- Process, filesystem, persistence, binary, and network string surfaces, what the code reaches for
- Phishing and brand-impersonation signals, shortlink funnels, lookalike domains, scam ad/tracking stacks, credential-harvest and wallet-drainer wording, and exfiltration endpoints (built from a real 2026 campaign that mass-mentioned GitHub users and redirected them to a fake
hermes-agent.icu) - Self-declared provenance and reuse signals, is this actually what it claims to be?
- AI/component inventory and agent/MCP workflow surfaces, repo code that drives agents
Public conclusions are deliberately bounded to:
No signal detected by this scanReviewRiskBlocked
Requirements: Python 3.11, 3.12, or 3.13.
Windows:
OPEN - MAYA Sentinel.cmd
Any supported platform:
python maya_lens_server.pyOpen http://127.0.0.1:5182/ and choose a repository ZIP you are authorized to inspect.
No dependency installation is required.
python maya_lens_server.py --scan path/to/repository.zip
python maya_lens_server.py --scan path/to/repository.zip --mode supply-chainModes: safety (default), phishing, supply-chain, ai-surface, exfil, archive.
AI-generated and AI-agent-driven code is everywhere now, and so is the temptation to install first, inspect never. Sentinel is the five-second inspection layer: bounded, local, and honest about what it can and cannot prove. The repo ecosystem is getting faster. Your review process should be, too.
Every referenced test ships in this repository:
python -m py_compile src/maya_lens/*.py maya_lens_server.py
python tests/test_maya_lens_scanner.py
python tests/test_maya_lens_server.py
python tests/test_public_release_contract.py
node --check web/app.jsA ready-to-enable GitHub Actions template is included at docs/ci/verify.yml.example; local verification remains the release authority.
- Uploaded ZIPs remain local and are removed after each scan attempt
- Raw scan state is memory-only by default
- Only public-projected reports and history metadata are retained locally
- Retained history and reports can be deleted through the UI
- The tool makes no repository network calls and sends no telemetry
The server binds to loopback and uses Host, Origin, and in-memory session-token checks for mutating requests. It emits CSP, anti-frame, no-sniff, referrer, permissions, COOP, and CORP browser hardening headers.
MIT License. See LICENSE.txt for the full terms.
Copyright (c) 2026 2ndNatureAi.