22 frameworks. One command. Mathematical proof that your agent skills are safe. Not just scanning — formal verification with soundness guarantees.
$ pip install skillfortify $ skillfortify scan SkillFortify is the type system for agent skills.
Not heuristics. Not pattern matching. Formal verification with five mathematical soundness theorems.
$ skillfortify scan Auto-discover and analyze every AI agent skill on your system. No arguments needed — finds Claude Code, Cursor, VS Code, and 19 more automatically.
$ skillfortify verify ./skills/deploy.md Formally verify a single skill against its capability declaration. Returns a detailed report with findings categorized by severity.
$ skillfortify lock ./my-agent
Generate skill-lock.json for reproducible, auditable agent configurations across environments.
$ skillfortify trust ./skills/deploy.md Compute formal trust scores with provenance, behavioral analysis, and community signals. Trust propagates through dependency chains.
$ skillfortify sbom ./my-agent Generate CycloneDX 1.6 Agent Skill Bill of Materials. Lists every skill, version, capabilities, dependencies, and trust level.
$ skillfortify frameworks List all 22 supported frameworks and their detection status on your system. Shows installed locations and skill counts.
$ skillfortify dashboard ./my-agent Generate a standalone HTML security dashboard. Interactive report with filters, risk charts, and capabilities matrix. Share as one file.
$ skillfortify registry-scan --source mcp Scan MCP registry, PyPI, and npm for malicious skills before you install them. Catch threats before they reach your system.
$ skillfortify verify ./skills/ --recursive Recursively verify every skill in a directory tree. Parallel analysis with aggregated results. Ideal for CI/CD pipelines.
Generate a complete security report as a standalone HTML file. No server. No SaaS subscription. Just open it in your browser.
$ skillfortify dashboard ./my-agent Share the HTML file with your security team. They don't need to install anything. One file tells the full story of your agent supply chain security posture.
+ auto-discovery of unknown tools — new framework released yesterday? SkillFortify finds it across 23+ IDE profiles without any update needed.
| Feature | SkillFortify | Snyk agent-scan | Cisco skill-scanner | Other Scanners |
|---|---|---|---|---|
| Formal Verification | ||||
| Soundness Theorems | 5 theorems | None | None | None |
| False Positive Rate | 0% | Not published | Not published | Varies |
| Agent Frameworks | 22 | Unknown | 2-3 | 1-2 |
| System Auto-Discovery | ||||
| HTML Dashboard | ||||
| Dependency Graph (SAT) | Partial | |||
| Lockfile Generation | ||||
| Trust Score Algebra | ||||
| CycloneDX ASBOM | Partial | |||
| Registry Scanning | MCP, PyPI, npm | Unknown | ||
| Peer-Reviewed Paper | ||||
| Open Source | Some | |||
| "No findings ≠ no risk" caveat | No — soundness proven | N/A | Yes (stated in docs) | Yes |
Inspired by SLSA (Supply-chain Levels for Software Artifacts), adapted for the agent skills ecosystem.
Built for the regulatory frameworks governing AI in production.
Machine-readable Agent Software Bill of Materials for every project.
Drop into your pipeline. Share interactive HTML reports with your security team.
Varun Pratap Bhardwaj, 2026
DOI: 10.5281/zenodo.18787663
Read on ZenodoThe DY-Skill threat model captures all possible symbolic attacks on the skill supply chain. Any real-world attack maps to a sequence of operations in the model.
If SkillFortify reports no capability violations, the skill provably does not exceed its declared capabilities. No false negatives for capability-level threats.
A skill executing within SkillFortify's capability sandbox cannot acquire capabilities beyond those explicitly granted. Authority cannot be amplified.
If the dependency resolver finds a valid configuration, all version constraints, conflict constraints, and security bounds are satisfied simultaneously.
Adding positive evidence to a skill's trust assessment never decreases its trust score. Trust propagation through dependency chains preserves ordering.
Supply Chain Security
How to VERIFY and SECURE the agent supply chain. 22 frameworks. 1,818 tests. Zero false positives.
Requires Python 3.11+. No cloud dependencies. No API keys. Runs entirely on your machine.
$ pip install skillfortify Built by Varun Pratap Bhardwaj · Part of the AgentAssert Research Suite
Star on GitHub