@haani090/latticeauth

Agent identity you can prove, not trust.

Two things you can run right now — no signup, no backend, fully offline:

# What's the worst this agent can do if it's fully prompt-injected?
npx @haani090/latticeauth policy blast-radius ./policy.yaml

# Prove an exported audit log wasn't edited after the fact.
npx @haani090/latticeauth audit verify ./audit-export.json

blast-radius collapses your policy's allow − deny into the concrete worst case: reachable scopes, max token lifetime, per-day cap, and which output channels are exfiltration paths — severity-rated, so it fails CI on an over-broad policy before it ships.

audit verify independently recomputes a SHA-256 hash chain over the export and reports ✓ VERIFIED or ✗ TAMPERING DETECTED at the exact offending event. Zero runtime dependencies, no network, no secrets — the audit trail's trustworthiness does not depend on trusting Lattice.

Install

npx @haani090/latticeauth --help        # no install
npm i -g @haani090/latticeauth          # or global: `lattice --help`

Requires Node ≥ 20. Self-contained single binary (everything bundled).

Commands

| Command | What it does | |---|---| | lattice policy blast-radius <policy.yaml> | Worst-case capability set under prompt injection | | lattice audit verify <bundle.json> | Offline tamper-evidence check of an audit export | | lattice mcp wrap <target> | Wrap an MCP server with Lattice OBO identity | | lattice audit tail [agentId] | Stream audit events | | lattice login / status | Control-plane auth + health |

Apache-2.0.