feat(ffi): external-token Ed25519 sign callback for federation-identity mint (Android YubiKey/NFC)

[emooreatx]

What

Adds ciris_verify_create_federation_identity_with_callback + a CallbackHardwareSigner (impl HardwareSigner) so a YubiKey-backed federation identity can be minted when the token is reachable only by an app's native layer (e.g. YubiKit over NFC on Android), not by this library.

Why

On desktop the node opens the YubiKey locally via ykcs11. On a phone the runtime is sandboxed and can't reach an NFC/USB token — the hardware ops must run in the app (YubiKit). The existing ciris_verify_create_federation_identity builds its own signer internally (get_platform_ed25519_signer) with no way to inject an app-provided one. This adds that seam.

How

The core composes the self_key_record + the platform-sealed ML-DSA-65 half exactly as create_federation_identity does, and delegates only the one classical Ed25519 signature to a caller-supplied C callback:

int ciris_verify_create_federation_identity_with_callback(
    const char* config_json,
    const uint8_t* ed25519_pubkey, size_t pubkey_len,   // 32, pre-read from slot 9c
    const uint8_t* attestation_der, size_t attestation_len, // optional slot-9c PIV attest
    void* sign_ctx, FfiEd25519SignCallback sign_cb,     // delegate (e.g. YubiKit session)
    char** result_out);

CallbackHardwareSigner::sign() invokes sign_cb synchronously; public_key() returns the passed-in key; attestation() reports the external-token tier. The app pre-reads the pubkey + attestation, so the core never touches the token.

Testing

• New unit tests (callback_signer::tests): delegation plumbing + callback-failure path — both green.
• cargo build -p ciris-verify-ffi + cargo test -p ciris-verify-ffi pass; pre-commit (fmt/clippy/doc/version) green.
• Not device-tested against a real YubiKey yet (no token/NFC in CI) — that lands with the Android client wiring in the 0.6.x cycle.

Scope / notes

• Additive: a new FFI symbol + private module, no breaking changes. Builds on the v6.13.0 ciris-keyring pkcs11: open() looks up the PUBLIC key with the PRIVATE key's class-specific label → KeyNotFound on YubiKey PIV (Ed25519 slot 9c) #112/accord custody attestation feeds the YubiKey Ed25519 a multi-KB preimage → C_Sign EdDSA CKR_DATA_LEN_RANGE (embeds full PIV attestation chain inline) #113 fixes.
• The C-ABI fn is callable today from the Python agent (Chaquopy ctypes). The JNI wrapper (Java_…_nativeCreateFederationIdentityWithCallback, which bridges sign_cb back into a Kotlin/YubiKit signEd25519) is the remaining Android glue — deferred to device testing. Client-side scaffolding (KMP YubiKeyPivSigner expect/actual) lives in CIRISServer.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[emooreatx]

Rebased onto v8.3.0 main (was ~5 releases stale, cut against v6.13.0-era main).

• The prior Test (windows-latest) failure was not a code bug — the Build step hit the crates.io HTTP/2 flake (download of cryptoki failed … [55] Failed receiving HTTP2 data). That flake is hardened on main since ci: harden crates.io fetch against the recurring HTTP/2 SSL-eof flake #148 (CARGO_NET_RETRY / CARGO_HTTP_MULTIPLEXING=false), which this branch now inherits.
• Rebase was clean — no conflicts (the new callback_signer.rs + the lib.rs module registration slotted in around the v7/v8 additions).
• Validated on the new base: cargo build -p ciris-verify-ffi ✓, 77 ffi tests pass (incl. callback_signer::tests::{delegates_sign_and_reports_pubkey, surfaces_callback_failure}), clippy -D warnings clean, fmt clean.

Mergeable again. Fresh CI running.