[zahin-mohammad]

What

• Upgrade the pinned Claude action to v1.0.260 and select the active Sonnet 5.5 global Bedrock inference profile with ANTHROPIC_MODEL in both Claude paths.
• Migrate bare @claude reviews to the supported prompt input and allow the GitHub CLI tools needed to read and post reviews.
• Make the review commands target the triggering repository and pull request explicitly.
• Keep custom mentions in interactive mode with comment and conversation context.
• Disable extended thinking because Bedrock rejects replayed tool-call turns without a leading thinking block.
• Replace the ignored timeout_minutes input with a ten-minute job timeout.
• Leave AWS roles, region, permissions, and application code unchanged.

Why

Claude reviews fail because the previous model reached end of life. The old action inputs are ignored, and comment-triggered runs start from the default branch, so the workflow also needs supported inputs and explicit pull-request targeting.

The linked BitGoJS failure showed that action v1.0.194 carried Claude Code 2.1.234, while Sonnet 5.5 requires 2.1.280 or newer. Action v1.0.260 carries Claude Agent SDK 0.3.309 and a sufficiently new Claude Code runtime.

Test plan

• actionlint -shellcheck= .github/workflows/claude-code.yaml passed using the BitGoJS locked Nix tool source.
• Targeted harness verified action v1.0.260, the global Sonnet 5.5 profile, MAX_THINKING_TOKENS=0, the ten-minute timeout, explicit GitHub CLI targeting, removal of obsolete inputs, and prompt selection for bare, whitespace-padded, and custom mentions.
• Verified prompt-versus-interactive mode behavior and accepted inputs against the pinned action source.
• Verified the pinned action carries Claude Agent SDK 0.3.309.
• After merge, invoke a bare @claude review and a custom mention; confirm successful inference and posted feedback. Comment-triggered workflows use the default branch definition, so live end-to-end verification cannot run from this PR branch.

Application builds and unit tests are not applicable to this workflow-only change. The inference role must allow the global Sonnet 5.5 profile and its destination models.

Ticket: WAL-2202

[linear-code]

WAL-2202

[zahin-mohammad]

AI-generated review — updated after validating the linked Sonnet 5.5 failure. Please use your own judgment before acting on any findings.

---

PR Review: #297 — restore Claude reviews with Sonnet 5.5

Changes: +17 / -9 in 1 file

Summary

This PR upgrades both review modes to claude-code-action v1.0.260, whose Claude Agent software development kit (SDK) dependency is 0.3.309, and selects the global Sonnet 5.5 Bedrock inference profile (.github/workflows/claude-code.yaml:84, .github/workflows/claude-code.yaml:87, .github/workflows/claude-code.yaml:107, .github/workflows/claude-code.yaml:110).

flowchart TD
    A[Owner or member comments @claude] --> B{Bare mention?}
    B -->|Yes| C[Load prompts/code-review.md]
    B -->|No| D[Preserve conversation context]
    C --> E[claude-code-action v1.0.260]
    D --> E
    E --> F[Global Bedrock Sonnet 5.5]
    F --> G[Post review to triggering pull request]

Loading

Findings

• ✅ Fixed: action v1.0.194 carried Claude Code 2.1.234, which the linked run showed cannot invoke Sonnet 5.5. The new pin carries SDK 0.3.309.
• ✅ Fixed: the profile now uses AWS's global Sonnet 5.5 identifier instead of the US-only profile (.github/workflows/claude-code.yaml:87).
• ✅ Fixed: review commands explicitly target the triggering repository and PR (.github/workflows/claude-code.yaml:101).
• ✅ Aligned with WAL-2202: bare reviews and interactive mentions retain their intended modes.

Verification

actionlint: PASS
latest-action/profile harness: PASS
commitlint: 0 errors
Mermaid render: PASS (202490 bytes)

Remaining live check: comment-triggered workflows use the default-branch definition, so actual Bedrock inference must be verified after merge. The inference role must permit the global profile and its worldwide destination-model resources.

---

Generated by the wallet-platform:review-pr skill

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.

_{Tip: disable this comment in your organization's Code Review settings.}

[zahin-mohammad]

@claude review always

[github-actions]

Claude encountered an error —— View job

---

I'll analyze this and get back to you.

[claude]

Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.

Additional findings (outside the current diff — GitHub can't attach inline comments there):

• 🟣 .github/workflows/claude-code.yaml — Pre-existing: an @ claude comment with an apostrophe or unmatched quote fails the whole job, so the custom-mention path this PR restores never runs. The Determine prompt to use step trims with echo "$COMMENT_BODY" | xargs. GNU xargs exits 1 with "unmatched single quote" on text like "why don't we retry?". The default bash -e shell then stops the step on that assignment. Fix: trim without xargs (for example TRIMMED_COMMENT=$(printf '%s' "$COMMENT_BODY" | tr -d '[:space:]') and compare to @ claude), so any comment text gets through. Bare and whitespace-padded @ claude must still pick the review prompt.

  Why this was flagged

  An OWNER or MEMBER posts an issue_comment or pull_request_review_comment such as "@ claude why don't we retry here?". The comment passes the job if: and reaches the Determine prompt to use step (.github/workflows/claude-code.yaml:56-62). There TRIMMED_COMMENT=$(echo "$COMMENT_BODY" | xargs) runs. GNU xargs treats quotes as special and exits 1 on an unmatched ' or ". A run: step's default shell is bash -e {0}, so a failing command substitution in an assignment ends the step. The job fails and neither claude-code-action step at :84 or :106 runs, so the mention gets no reply. The same happens for any comment with an odd number of quotes. This is pre-existing: the base fails the same way at this line. This PR does not touch the step, but it says it restores custom mentions in interactive mode (:106-117), and those runs depend on this step's use-code-review-prompt=false output. That path stays broken for ordinary English comments. The test plan says custom mentions were checked, but that check could not have used a comment with an apostrophe.

  Verification: Line 65 of the Determine prompt to use step runs TRIMMED_COMMENT=$(echo "$COMMENT_BODY" | xargs). If the text has an unmatched quote, GNU xargs exits 1. The step uses the default bash -e {0}, so it aborts and use-code-review-prompt is never set. Both action steps (lines 85 and 108) are skipped. This step is byte-for-byte unchanged by the diff, so the base branch fails the same way.