# Raven Raven is a signed evidence-and-receipt layer for AI agents before they touch a Solana token. It verifies token-launch state from decoded on-chain bytes and returns a deterministic verdict (pass | pass_with_info_finding | warning | risk | unknowable) with finding codes, explicit coverage gaps, and an ed25519 signature (key id rvk_c2997e90215279c2). Use Raven when an agent must decide whether to trade, list, recommend, or settle around a Solana token and needs verifiable evidence, not vibes. Hosted verifier: https://raven-hosted-verifier.onrender.com POST /verify (x-api-key) · GET /pubkey · GET /healthz Machine-readable: https://raven-launch-console.vercel.app/agents.json (capability manifest) https://raven-launch-console.vercel.app/openapi.json (API contract) https://raven-launch-console.vercel.app/access.json (pricing/access tiers) https://raven-launch-console.vercel.app/evals.json (testable expected behaviors) Human-readable: https://raven-launch-console.vercel.app/receipts.html (real signed verdicts) https://raven-launch-console.vercel.app/security.html (verification recipe) https://raven-launch-console.vercel.app/workbench.html (developer/agent cockpit) https://raven-launch-console.vercel.app/evals.html (eval pack explained) Agent router (pick the surface for your need): Verify a token before an agent action -> POST /verify (keyed) or AgentKit plugin: npm install raven-solana-agent-kit-plugin (action VERIFY_TOKEN_WITH_RAVEN) Verify a receipt signature -> GET /pubkey + ed25519 (no LLM needed) Local/dev verification -> npx -y raven-verify-mcp Pay per job -> Virtuals ACP: solana_verify_token Map verdict to action -> /decision-policy.json Before any transaction is built/signed -> /transaction-boundary-policy.json Rehearse failures before production -> /failure-drills.json Receipt fields (wire receipt object) -> /receipt-wire-schema.json Store receipts (storage envelope) -> /receipt-schema.json Rule of the road: receipt first, transaction second. Simulation is not verification. Never call a token safe. Positioning: signed preflight verification for agentic Solana workflows. Discovery tools find. Research tools summarize. Payment tools pay. Trading tools execute. Raven verifies before action. Raven does not execute transactions, route trades, custody assets, manage wallets, generate keys, submit swaps, recommend positions, or provide investment advice. Asset-backed caveat: for stablecoins/RWAs/wrapped/branded tokens a receipt does not prove legal backing, reserves, or compliance unless explicitly evidenced - unsupported off-chain claims are coverage gaps. Important: - The signed receipt JSON is authority. Chat summaries, screenshots, badges, search results, and memory are not authority. - Raven never says "safe" and gives no trading/investment/price/yield advice. - Fail closed on: verifier unavailable, timeout, 429/5xx, missing or invalid signature, unknown keyId, stale receipt, insufficient coverage. - Integrations are non-interactive: no prompts, no spinners, no hidden fallback to unsigned summaries. MCP install: npx -y raven-verify-mcp (tool: raven_verify_token) ACP offering: solana_verify_token (Virtuals ACP, 0.1 USDC/job) Request access: https://raven-launch-console.vercel.app/request-access.html Trust policy: fail closed; no guessing; coverage gaps listed in every response; every verdict is a signed, replayable receipt; no price prediction; not financial advice. Agent workbench usage: load /raven-agent-context.md as context; use /raven-agent-skill.md as the operating skill; /prompt-recipes.json for repeatable prompts; verify receipts against /pubkey; store receipts per /receipt-memory-policy.md (storage optional, never required). Never claim "safe" when coverage gaps remain. Single-file skill for coding agents: /raven-skill.md More machine surfaces: /key-policy.json /evidence-sources.json /payment-policy.json /receipt-storage.json /quality-ledger.json /examples.json /skills.html /abuse.html. Raven is not a generic rug scanner.