[ci] feat: add @claude PR review workflow via claude-code-action

Add a comment-triggered workflow that runs Claude Code as a PR review
agent, using the upstream anthropics/claude-code-action@v1.

How it works:
- Trigger: comments matching '@claude' on issues, PRs, PR reviews, and
  PR review comments. Default phrase per upstream action.
- Authorization gate: only OWNER / MEMBER / COLLABORATOR commenters
  can fire the action, with a self-author bypass for the original
  PR/issue author. Without this gate, drive-by commenters on a public
  repo could trigger paid agent runs and execute tools on the
  self-hosted runner.
- Permissions: contents: read (agent reads files only),
  pull-requests: write + issues: write (agent posts review comments),
  id-token: write (action's OIDC token request),
  actions: read.
- Routing: env.ANTHROPIC_BASE_URL routes Claude API traffic through
  the internal gateway. Verified end to end via the Claude Code CLI;
  the gateway supports tool use + streaming.
- Model: 'Claude Sonnet 4.6' — the gateway exposes Claude models with
  display-name IDs (verified via GET /v1/models). The canonical
  hyphenated form is rejected as 'no available channel'.
- Tools: --allowedTools enables the inline review-comment MCP tool
  plus the gh / Read / Grep / Glob tools the agent needs to inspect
  the diff and modified files.
- Behavior: prompt is intentionally empty so the action runs in its
  default agent mode (PR context + GitHub MCP tools auto-injected).
  Project-wide review guidance lives in CLAUDE.md and
  skills/code-review/SKILL.md, which the agent loads automatically.
- Progress: track_progress: true shows live checkbox progress in the
  bot's initial PR comment.

End-to-end verified on a fork PR: the agent posts severity-tagged
inline comments on file:line plus a top-level verdict / summary /
checklist, matching the SKILL.md contract.

Author: Zachary-wW

.github/workflows/claude-review.yml

@@ -0,0 +1,96 @@
+name: Claude Code
+
+on:
+  issue_comment:
+    types: [created]
+  pull_request_review_comment:
+    types: [created]
+  pull_request_review:
+    types: [submitted]
+  issues:
+    types: [opened, assigned]
+
+permissions:
+  contents: read
+  pull-requests: write
+  issues: write
+  actions: read
+  id-token: write
+
+jobs:
+  claude:
+    # Run only when @claude is mentioned AND the commenter is a trusted
+    # actor on the repo (or the PR/issue author themselves). Without this
+    # gate, anyone on the public internet who can comment on the repo
+    # (e.g. a fork-PR contributor) could trigger paid agent runs and
+    # execute tools on the self-hosted runner.
+    if: >-
+      (
+        github.event_name == 'issue_comment' &&
+        contains(github.event.comment.body, '@claude') &&
+        (
+          contains(fromJSON('["OWNER","MEMBER","COLLABORATOR"]'), github.event.comment.author_association) ||
+          github.event.comment.user.login == github.event.issue.user.login
+        )
+      ) ||
+      (
+        github.event_name == 'pull_request_review_comment' &&
+        contains(github.event.comment.body, '@claude') &&
+        (
+          contains(fromJSON('["OWNER","MEMBER","COLLABORATOR"]'), github.event.comment.author_association) ||
+          github.event.comment.user.login == github.event.pull_request.user.login
+        )
+      ) ||
+      (
+        github.event_name == 'pull_request_review' &&
+        contains(github.event.review.body, '@claude') &&
+        (
+          contains(fromJSON('["OWNER","MEMBER","COLLABORATOR"]'), github.event.review.author_association) ||
+          github.event.review.user.login == github.event.pull_request.user.login
+        )
+      ) ||
+      (
+        github.event_name == 'issues' &&
+        (contains(github.event.issue.body, '@claude') || contains(github.event.issue.title, '@claude')) &&
+        contains(fromJSON('["OWNER","MEMBER","COLLABORATOR"]'), github.event.issue.author_association)
+      )
+    runs-on: [self-hosted, macOS, ARM64]
+    timeout-minutes: 30
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Claude Code Action Official
+        uses: anthropics/claude-code-action@v1
+        env:
+          # Route Claude API traffic through the internal proxy.
+          ANTHROPIC_BASE_URL: ${{ secrets.ANTHROPIC_BASE_URL }}
+        with:
+          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
+          # track_progress shows a live checkbox list in the bot's
+          # initial PR comment as the agent works.
+          track_progress: true
+          # Default trigger phrase is @claude.
+          # claude_args:
+          # - Model ID matches what the internal gateway exposes via
+          #   /v1/models (display-name form with spaces, not the
+          #   canonical Anthropic ID).
+          # - --allowedTools explicitly enables the inline review-comment
+          #   MCP tool (not in the default allow-list) plus the gh
+          #   subcommands and read-only filesystem tools the agent needs
+          #   to actually inspect the diff and modified files.
+          claude_args: |
+            --model "Claude Sonnet 4.6"
+            --max-turns 20
+            --allowedTools "mcp__github_inline_comment__create_inline_comment,Bash(gh pr comment:*),Bash(gh pr diff:*),Bash(gh pr view:*),Read,Grep,Glob"
+          # NOTE: do NOT set `prompt:` here. Setting it puts the action in
+          # custom-automation mode, which overrides the user's @claude
+          # comment and skips the GitHub MCP tool injection — meaning the
+          # agent has no way to post a review back. With `prompt:` empty,
+          # the action's default agent mode injects PR context + GitHub
+          # tools, and the user's @claude comment becomes the agent input.
+          # Project-wide instructions (e.g. "use skills/code-review/SKILL.md
+          # when reviewing") belong in CLAUDE.md, which the agent loads
+          # automatically.