feat(auth): per-sandbox authentication to gateway

.agents/skills/debug-openshell-cluster/SKILL.md

@@ -116,9 +116,16 @@ Check required Helm deployment secrets:
 kubectl -n openshell get secret \
   openshell-server-tls \
   openshell-server-client-ca \
-  openshell-client-tls
+  openshell-client-tls \
+  openshell-jwt-keys
 ```
 
+If the gateway exits with `failed to read sandbox JWT signing key from
+/etc/openshell-jwt/signing.pem`, verify that `openshell-jwt-keys` contains
+`signing.pem`, `public.pem`, and `kid`, and that the StatefulSet mounts the
+`sandbox-jwt` secret at `/etc/openshell-jwt`. The sandbox JWT mount is required
+even when local Helm values disable TLS.
+
 Check the image references currently used by the gateway deployment:
 
 ```bash

crates/openshell-bootstrap/Cargo.toml

@@ -16,6 +16,7 @@ bytes = { workspace = true }
 futures = { workspace = true }
 miette = { workspace = true }
 rcgen = { workspace = true }
+sha2 = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }
 tar = "0.4"

crates/openshell-bootstrap/src/jwt.rs

@@ -0,0 +1,112 @@
+// SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+//! Gateway-minted JWT signing-key generation.
+//!
+//! The gateway mints per-sandbox identity tokens (see PR 2 of the
+//! per-sandbox identity series, issue #1354) signed with an Ed25519
+//! keypair generated once at gateway init and persisted alongside the
+//! existing PKI bundle. The signing key never leaves the gateway; the
+//! public key plus a stable `kid` are consumed by the gateway's own
+//! validator and any future external verifiers.
+
+use miette::{IntoDiagnostic, Result, WrapErr};
+use rcgen::{KeyPair, PKCS_ED25519};
+use sha2::{Digest, Sha256};
+
+/// All PEM-encoded material needed to mint and validate sandbox JWTs.
+///
+/// The signing key stays in the gateway process. The public key is shared
+/// across gateway replicas (so any replica can validate a JWT minted by
+/// any other replica). The `kid` is published in every minted JWT's
+/// header so the validator can pick the right key after a future rotation.
+pub struct JwtKeyMaterial {
+    /// PKCS#8 PEM-encoded Ed25519 private key.
+    pub signing_key_pem: String,
+    /// `SubjectPublicKeyInfo` PEM-encoded Ed25519 public key.
+    pub public_key_pem: String,
+    /// Stable identifier derived from the public key (SHA-256 hex prefix).
+    /// Embedded in every minted JWT's `kid` header so future rotation can
+    /// be performed in-place by adding a second key without breaking
+    /// in-flight tokens.
+    pub kid: String,
+}
+
+/// Generate a fresh Ed25519 JWT signing key.
+///
+/// Output PEM is in the formats `jsonwebtoken` consumes via
+/// `EncodingKey::from_ed_pem` (signing) and `DecodingKey::from_ed_pem`
+/// (validation), so the gateway can round-trip its own tokens with no
+/// further conversion.
+pub fn generate_jwt_key() -> Result<JwtKeyMaterial> {
+    let keypair = KeyPair::generate_for(&PKCS_ED25519)
+        .into_diagnostic()
+        .wrap_err("failed to generate Ed25519 JWT signing key")?;
+    let signing_key_pem = keypair.serialize_pem();
+    let public_key_pem = keypair.public_key_pem();
+    let kid = kid_from_public_key_der(&keypair.public_key_der());
+    Ok(JwtKeyMaterial {
+        signing_key_pem,
+        public_key_pem,
+        kid,
+    })
+}
+
+/// Stable `kid` derived from the SHA-256 of the public-key DER.
+///
+/// First 16 bytes hex-encoded — collision-resistant for the small N of
+/// signing keys a single deployment ever has, while staying short enough
+/// to keep JWT headers compact.
+fn kid_from_public_key_der(public_key_der: &[u8]) -> String {
+    let digest = Sha256::digest(public_key_der);
+    hex_encode_prefix(&digest, 16)
+}
+
+fn hex_encode_prefix(bytes: &[u8], n: usize) -> String {
+    use std::fmt::Write as _;
+    let mut out = String::with_capacity(n * 2);
+    for byte in bytes.iter().take(n) {
+        let _ = write!(out, "{byte:02x}");
+    }
+    out
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn generate_jwt_key_produces_parseable_pem() {
+        let material = generate_jwt_key().expect("generate_jwt_key");
+        assert!(material.signing_key_pem.contains("BEGIN PRIVATE KEY"));
+        assert!(material.public_key_pem.contains("BEGIN PUBLIC KEY"));
+        assert_eq!(material.kid.len(), 32, "kid is 16 bytes hex-encoded");
+        assert!(material.kid.chars().all(|c| c.is_ascii_hexdigit()));
+    }
+
+    #[test]
+    fn kid_is_stable_for_identical_public_keys() {
+        // Same input -> same kid. Hash of a fixed byte string.
+        let kid_a = kid_from_public_key_der(b"abc");
+        let kid_b = kid_from_public_key_der(b"abc");
+        assert_eq!(kid_a, kid_b);
+    }
+
+    #[test]
+    fn kid_differs_for_different_public_keys() {
+        let kid_a = kid_from_public_key_der(b"first");
+        let kid_b = kid_from_public_key_der(b"second");
+        assert_ne!(kid_a, kid_b);
+    }
+
+    #[test]
+    fn generated_keys_are_unique() {
+        let a = generate_jwt_key().expect("generate_jwt_key");
+        let b = generate_jwt_key().expect("generate_jwt_key");
+        assert_ne!(
+            a.kid, b.kid,
+            "fresh keypairs must produce distinct public keys"
+        );
+        assert_ne!(a.signing_key_pem, b.signing_key_pem);
+    }
+}

crates/openshell-bootstrap/src/lib.rs

@@ -3,6 +3,7 @@
 
 pub mod build;
 pub mod edge_token;
+pub mod jwt;
 pub mod oidc_token;
 
 mod metadata;

crates/openshell-bootstrap/src/pki.rs

@@ -1,6 +1,7 @@
 // SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 // SPDX-License-Identifier: Apache-2.0
 
+use crate::jwt::{JwtKeyMaterial, generate_jwt_key};
 use miette::{IntoDiagnostic, Result, WrapErr};
 use rcgen::{BasicConstraints, CertificateParams, DnType, Ia5String, IsCa, KeyPair, SanType};
 use std::net::IpAddr;
@@ -15,6 +16,12 @@ pub struct PkiBundle {
     pub server_key_pem: String,
     pub client_cert_pem: String,
     pub client_key_pem: String,
+    /// PKCS#8 PEM Ed25519 private key for minting per-sandbox JWTs.
+    pub jwt_signing_key_pem: String,
+    /// SPKI PEM Ed25519 public key, paired with `jwt_signing_key_pem`.
+    pub jwt_public_key_pem: String,
+    /// Stable identifier embedded in the `kid` header of every minted JWT.
+    pub jwt_key_id: String,
 }
 
 /// Default SANs always included on the server certificate.
@@ -95,13 +102,23 @@ pub fn generate_pki(extra_sans: &[String]) -> Result<PkiBundle> {
         .into_diagnostic()
         .wrap_err("failed to sign client certificate")?;
 
+    // --- JWT signing key (Ed25519, used to mint per-sandbox identity tokens) ---
+    let JwtKeyMaterial {
+        signing_key_pem: jwt_signing_key_pem,
+        public_key_pem: jwt_public_key_pem,
+        kid: jwt_key_id,
+    } = generate_jwt_key().wrap_err("failed to generate JWT signing key")?;
+
     Ok(PkiBundle {
         ca_cert_pem: ca_cert.pem(),
         ca_key_pem: ca_key.serialize_pem(),
         server_cert_pem: server_cert.pem(),
         server_key_pem: server_key.serialize_pem(),
         client_cert_pem: client_cert.pem(),
         client_key_pem: client_key.serialize_pem(),
+        jwt_signing_key_pem,
+        jwt_public_key_pem,
+        jwt_key_id,
     })
 }
 
@@ -144,6 +161,9 @@ mod tests {
         assert!(bundle.server_key_pem.contains("BEGIN PRIVATE KEY"));
         assert!(bundle.client_cert_pem.contains("BEGIN CERTIFICATE"));
         assert!(bundle.client_key_pem.contains("BEGIN PRIVATE KEY"));
+        assert!(bundle.jwt_signing_key_pem.contains("BEGIN PRIVATE KEY"));
+        assert!(bundle.jwt_public_key_pem.contains("BEGIN PUBLIC KEY"));
+        assert_eq!(bundle.jwt_key_id.len(), 32, "kid is 16 bytes hex-encoded");
     }
 
     #[test]

crates/openshell-cli/src/run.rs

@@ -739,6 +739,11 @@ fn import_local_package_mtls_bundle(name: &str) -> Result<Option<PathBuf>> {
             client_key_pem: std::fs::read_to_string(&key)
                 .into_diagnostic()
                 .wrap_err_with(|| format!("failed to read {}", key.display()))?,
+            // CLI never holds the gateway's JWT signing material — only the
+            // gateway needs it. Fill the JWT fields with placeholders.
+            jwt_signing_key_pem: String::new(),
+            jwt_public_key_pem: String::new(),
+            jwt_key_id: String::new(),
         };
         openshell_bootstrap::mtls::store_pki_bundle(name, &bundle)
             .wrap_err_with(|| format!("failed to store mTLS bundle for gateway '{name}'"))?;

crates/openshell-cli/tests/ensure_providers_integration.rs

@@ -488,6 +488,20 @@ impl OpenShell for TestOpenShell {
         Err(Status::unimplemented("not implemented in test"))
     }
 
+    async fn issue_sandbox_token(
+        &self,
+        _request: tonic::Request<openshell_core::proto::IssueSandboxTokenRequest>,
+    ) -> Result<Response<openshell_core::proto::IssueSandboxTokenResponse>, Status> {
+        Err(Status::unimplemented("not implemented in test"))
+    }
+
+    async fn refresh_sandbox_token(
+        &self,
+        _request: tonic::Request<openshell_core::proto::RefreshSandboxTokenRequest>,
+    ) -> Result<Response<openshell_core::proto::RefreshSandboxTokenResponse>, Status> {
+        Err(Status::unimplemented("not implemented in test"))
+    }
+
     async fn connect_supervisor(
         &self,
         _request: tonic::Request<tonic::Streaming<SupervisorMessage>>,

crates/openshell-cli/tests/mtls_integration.rs

@@ -397,6 +397,20 @@ impl OpenShell for TestOpenShell {
         Err(Status::unimplemented("not implemented in test"))
     }
 
+    async fn issue_sandbox_token(
+        &self,
+        _request: tonic::Request<openshell_core::proto::IssueSandboxTokenRequest>,
+    ) -> Result<Response<openshell_core::proto::IssueSandboxTokenResponse>, Status> {
+        Err(Status::unimplemented("not implemented in test"))
+    }
+
+    async fn refresh_sandbox_token(
+        &self,
+        _request: tonic::Request<openshell_core::proto::RefreshSandboxTokenRequest>,
+    ) -> Result<Response<openshell_core::proto::RefreshSandboxTokenResponse>, Status> {
+        Err(Status::unimplemented("not implemented in test"))
+    }
+
     async fn connect_supervisor(
         &self,
         _request: tonic::Request<tonic::Streaming<openshell_core::proto::SupervisorMessage>>,

crates/openshell-cli/tests/provider_commands_integration.rs

@@ -620,6 +620,20 @@ impl OpenShell for TestOpenShell {
         Err(Status::unimplemented("not implemented in test"))
     }
 
+    async fn issue_sandbox_token(
+        &self,
+        _request: tonic::Request<openshell_core::proto::IssueSandboxTokenRequest>,
+    ) -> Result<Response<openshell_core::proto::IssueSandboxTokenResponse>, Status> {
+        Err(Status::unimplemented("not implemented in test"))
+    }
+
+    async fn refresh_sandbox_token(
+        &self,
+        _request: tonic::Request<openshell_core::proto::RefreshSandboxTokenRequest>,
+    ) -> Result<Response<openshell_core::proto::RefreshSandboxTokenResponse>, Status> {
+        Err(Status::unimplemented("not implemented in test"))
+    }
+
     async fn connect_supervisor(
         &self,
         _request: tonic::Request<tonic::Streaming<SupervisorMessage>>,

crates/openshell-cli/tests/sandbox_create_lifecycle_integration.rs

@@ -574,6 +574,20 @@ impl OpenShell for TestOpenShell {
         Err(Status::unimplemented("not implemented in test"))
     }
 
+    async fn issue_sandbox_token(
+        &self,
+        _request: tonic::Request<openshell_core::proto::IssueSandboxTokenRequest>,
+    ) -> Result<Response<openshell_core::proto::IssueSandboxTokenResponse>, Status> {
+        Err(Status::unimplemented("not implemented in test"))
+    }
+
+    async fn refresh_sandbox_token(
+        &self,
+        _request: tonic::Request<openshell_core::proto::RefreshSandboxTokenRequest>,
+    ) -> Result<Response<openshell_core::proto::RefreshSandboxTokenResponse>, Status> {
+        Err(Status::unimplemented("not implemented in test"))
+    }
+
     async fn connect_supervisor(
         &self,
         _request: tonic::Request<tonic::Streaming<SupervisorMessage>>,

crates/openshell-cli/tests/sandbox_name_fallback_integration.rs

@@ -409,6 +409,20 @@ impl OpenShell for TestOpenShell {
         Err(Status::unimplemented("not implemented in test"))
     }
 
+    async fn issue_sandbox_token(
+        &self,
+        _request: tonic::Request<openshell_core::proto::IssueSandboxTokenRequest>,
+    ) -> Result<Response<openshell_core::proto::IssueSandboxTokenResponse>, Status> {
+        Err(Status::unimplemented("not implemented in test"))
+    }
+
+    async fn refresh_sandbox_token(
+        &self,
+        _request: tonic::Request<openshell_core::proto::RefreshSandboxTokenRequest>,
+    ) -> Result<Response<openshell_core::proto::RefreshSandboxTokenResponse>, Status> {
+        Err(Status::unimplemented("not implemented in test"))
+    }
+
     async fn connect_supervisor(
         &self,
         _request: tonic::Request<tonic::Streaming<SupervisorMessage>>,