SkillProof Verification Tiers
Every audit tests behavioral invariants under adversarial pressure, producing an Ed25519-signed Trust Manifest pinned to your code hash.
SkillProof Sprint
Adversarial injection-resistance and claim verification battery for a single agent skill or MCP server.
What is verified:
- 180 adversarial ops across 5 categories
- Direct override, tool injection, encoding smuggling
- Ed25519-signed Trust Manifest
- Offline-verifiable cryptographic proof
- CI regression integration snippet
MCP Verification
Full stack verification for MCP server providers and marketplace listings (Smithery, Glama).
What is verified:
- Identity & OAuth credential leakage tests
- Tool schema poisoning & injection battery
- File and shell exposure audit
- Verified MCP marketplace badge embed
- 48-hour re-test window upon remediation
Agent Payments Permission Audit
Behavioral proof that your agent cannot move money, invoice, or transfer funds without human approval.
What is verified:
- 500+ adversarial payment manipulation operations
- Zero unauthorized payment invariant testing
- Idempotency and double-spend resistance
- OAuth & Stripe credential isolation audit
- Executive risk scorecard + Ed25519 Trust Manifest
VC / Architectural Diligence
Independent technical audit for pre-seed/seed investors and enterprise procurement officers evaluating AI agent companies.
What is verified:
- Full architecture & harness deconstruction
- Simulated multi-agent collusion & drift analysis
- Data exfiltration attack simulations
- Confidential LP diligence report + debrief call
Trust Manifest Invariant Standard
A pass means our batteries failed to break your agent — not that no vulnerability exists. All findings cite the exact reproducible op string. Unknowns are counted as failures. Never rely on vendor claim marketing when verifiable code proofs exist.