fix(bitcoin-adapter): initialize ECC lib for Taproot PSBT finalization

[pedromcunha]

Taproot key-path finalization requires an ECC library; without it bitcoinjs-lib throws 'No ECC Library provided'. Also skip inputs the wallet already finalized, which otherwise throw 'Can not finalize input #0' (OKX autoFinalized).

Fixes INT2-1684, FE-8536

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
relay-sdk-demo	Ready	Preview	Oct 5, 2026 7:48pm UTC

[github-actions]

Claude finished @pedromcunha's task in 25s —— View job

---

Code Review

• Gather context (diff, commits, full file)
• Review for correctness, security, error handling, repo guidelines

Code Review — No issues found.

Reviewed 4 files, 27 additions / 1 deletion. Checked for correctness, security, error handling, and repo guidelines compliance.

The fix is correct and well-scoped:

• bitcoin.initEccLib(ecc) before finalization resolves the No ECC Library provided throw for Taproot key-path inputs (@bitcoinerlab/secp256k1 is the standard ECC provider for bitcoinjs-lib v7, and its @noble/curves dependency aligns with v7's expectations).
• Guarding each input on !input.finalScriptSig && !input.finalScriptWitness before calling finalizeInput(index) correctly skips wallet-auto-finalized inputs (OKX), avoiding the Can not finalize input #0 error, while still finalizing the ones the SDK must finalize.
• Dependency pinned to an exact version (1.2.0), consistent with how bitcoinjs-lib is pinned in this package.
• Changeset is reader-facing and names the exact errors and surface — matches the AGENTS.md changeset guidance.

A couple of minor, non-blocking observations (not issues):

• bitcoin.initEccLib(ecc) is idempotent, so calling it per handleSendTransactionStep invocation is harmless; it could equally live at module scope. Current placement is fine.
• The comment on line 65 accurately describes the behavior and is evergreen.

[greptile-apps]

Confidence Score: 4/5

[High risk] Adds a cryptographic dependency to Bitcoin transaction signing.

The PR appears safe to merge, with non-blocking regression coverage needed for both wallet finalization paths.

Findings

1.  Finalization paths lack tests ▶

Summary

The PR adds an ECC implementation to the Bitcoin adapter and initializes it before finalization, while leaving wallet-finalized PSBT inputs untouched.

• These changes target UniSat Taproot and OKX pre-broadcast failures.
• Regression tests for the two changed paths are still needed.
• Greptile automatically discovered related tickets that helped explain the purpose of this PR: restoring UniSat Taproot finalization and handling OKX wallet-finalized inputs.

_{Reviews (1) · Last reviewed commit: "fix(bitcoin-adapter): initialize ECC lib..."}

[greptile-apps]

Finalization paths lack tests

There are no automated tests for the two paths changed here: finalizing a signed Taproot input after ECC initialization, and extracting a transaction when the wallet has already finalized an input. The adapter has no test script, and PR CI runs only the SDK tests, so a recurrence of either reported failure would go undetected. Greptile automatically discovered related tickets stating that both the UniSat Taproot and OKX finalized-input fixes need regression coverage, which informed this comment.

Source Used: Linear — UniSat Taproot swaps fail before broadcast
Source Used: Linear — OKX Taproot PSBTs fail during finalization

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[github-actions]

Claude finished @pedromcunha's task in 34s —— View job

---

📝 Docs Drift Check — No docs impact

This PR does not touch surfaces that require a relayprotocol/relay-docs update.