SkillFortify is the first formal analysis framework for agent skill supply chain security. The ClawHavoc campaign (1,200+ malicious skills), CVE-2026-25253, and 6,487 malicious tools catalogued by MalTool demonstrate that the current "install and trust" paradigm for AI agent skills is untenable. While 12+ reactive scanning tools emerged in February 2026, all rely on heuristic methods with no formal guarantees. SkillFortify provides five formal contributions: (1) the DY-Skill attacker model adapting Dolev-Yao to agent supply chains, (2) sound static analysis via abstract interpretation over a capability lattice, (3) capability confinement with formal proof, (4) SAT-based dependency resolution with lockfile semantics, and (5) trust score algebra with formal monotonicity. Evaluated on SkillFortifyBench (540 skills), the framework achieves F1=96.95% with 100% precision and 0% false positive rate. Tool: pip install skillfortify | Code: https://github.com/varun369/skillfortify