SVS | Solana Agent Authorization

Action authorization and verifiable execution evidence for Solana agents. Apply policy, bind approval to the exact transaction, and prove what executed.

[SVS | Solana Agent Authorization @SVSProtocol]

Solana agents move fast. Consequential actions need a control boundary.

SVS authenticates exact requests, applies policy, binds approval to transaction bytes, verifies execution, and leaves portable evidence.

Control before execution. Proof after execution.
https://t.co/edDbP75Jmv

[SVS | Solana Agent Authorization @SVSProtocol]

An approval must not authorize a near match.

SVS flips one serialized-transaction byte and requires rejection at “Exact serialized transaction” before broadcast. The probe ships in our public reference workflow: https://t.co/HOHIkuc5wr

[SVS | Solana Agent Authorization @SVSProtocol]

Treasury agents need a stricter acceptance rule: policy review, operator approval, exact transaction binding, then independently verifiable registry evidence.

Public SVS devnet demo:
https://t.co/ChiYU4CPb7

[SVS | Solana Agent Authorization @SVSProtocol]

Change one byte in approved transaction bytes.

The SVS workflow must reject it before broadcast with:
failureCode: action_authorization_failed
failedCheck: Exact serialized transaction

A generic denial does not pass.
https://t.co/HOHIkuc5wr

[SVS | Solana Agent Authorization @SVSProtocol]

Security claims need testable failure conditions.

SVS's public review packet targets unauthenticated requests, replayed nonces, stale approvals, altered transactions, and missing execution evidence.

No third-party audit claimed: https://t.co/5Dsh3HlpcV

[SVS | Solana Agent Authorization @SVSProtocol]

One external agent. One relying party. One consequential action.

The SVS reference workflow checks the exact transaction before broadcast, verifies execution afterward, and requires mismatch probes to fail.

Pilot starter: https://t.co/HOHIkuc5wr

[SVS | Solana Agent Authorization @SVSProtocol]

Identity answers who the agent is. It should not grant blanket execution rights.

@svsprotocol/[email protected] can bind an official Agent Registry asset’s operational wallet into requireAuthorizedAction, then reject a different wallet or transaction.

[SVS | Solana Agent Authorization @SVSProtocol]

A signed request should work once, not forever.

SVS replay proof:
1. Accept the authenticated action.
2. Persist its nonce.
3. Resubmit the same signed nonce.
4. Reject it with HTTP 409.

The rejection is stored as evidence, not assumed.

[SVS | Solana Agent Authorization @SVSProtocol]

npm install @svsprotocol/[email protected]

The new AgentFund x402 pilot adapter can require an exact-action match before relay, report external broadcast evidence, and verify post-execution acceptance.

This is a pilot surface, not an adopted integration.

[SVS | Solana Agent Authorization @SVSProtocol]

SVS does not ask protocols to trust a badge.

Paste a proof summary into the public verifier demo: it checks the verification-set version, registry hash, action-proof status, and trusted signer result.

Inspect it: https://t.co/sjxltt82ZJ

[SVS | Solana Agent Authorization @SVSProtocol]

A badge is not proof. SVS publishes a registry hash, trust manifest, linked profile hashes, freshness status, and a verification command so a protocol can independently reject stale or mismatched agent claims.
Inspect the devnet demo registry: https://t.co/oePQB9wEUZ

[SVS | Solana Agent Authorization @SVSProtocol]

AgentFund reviewed the open SVS x402 settlement-gate PR and requested a vendor-neutral provider interface, exact SDK pinning, and a verified-contribution event. The changes are now in the PR; enforcement remains off pending review. https://t.co/QQH0Qp2NXT

[SVS | Solana Agent Authorization @SVSProtocol]

SVS is looking for one narrow reference workflow:
• one external Solana agent
• one safe devnet action
• one protocol, wallet, or operator enforcing the gate
• one sanitized result
The starter and pilot proposal are public:
https://t.co/GHjFVZdThG

[SVS | Solana Agent Authorization @SVSProtocol]

Security reviewers should not have to take an agent system on faith.

SVS publishes a review packet covering the SDK, two-phase workflow, registry and trust manifest, verifier guidance, threat model, and deliberate mismatch probes.

https://t.co/5Dsh3HkRnn

[SVS | Solana Agent Authorization @SVSProtocol]

New for protocol-owned relayers: reportExternalBroadcast(...)
Send an HMAC-signed transaction signature. SVS confirms it through the operator-pinned Solana RPC, matches the exact approved bytes, and rejects unconfirmed, foreign, or mismatched reports.

[SVS | Solana Agent Authorization @SVSProtocol]

One acceptance rule can change how protocols handle automation:

Accept an agent action only when the exact request passes policy, carries the required approval, preserves transaction integrity, and returns a fresh SVS verification result.

Start with one devnet action.

[SVS | Solana Agent Authorization @SVSProtocol]

Retries should not become duplicate agent actions. SVS reuses the exact signed body, timestamp, nonce, and HMAC only when a stable idempotency key is present. Reuse that key with changed action details and the request fails instead of quietly mutating intent.

[SVS | Solana Agent Authorization @SVSProtocol]

Agents can now install the SVS action-authorization skill:

npx skills add https://t.co/XEiO7V2Qyo --skill svs-action-authorization
It teaches fail-closed readiness, signed submission, and independent proof checks.