code-review plugin completes review but produces empty result — no PR comment posted

# Describe the bug
The `code-review@claude-code-plugins` plugin successfully performs a full code review (spawns parallel review agents, validates findings, outputs a formatted summary), but the session ends with `"result": ""`, so no comment is ever posted to the PR.

The root cause appears to be that the plugin's final turn is a `TodoWrite` tool call (marking its last checklist item as "completed") rather than a text message. The actual review summary was generated as a text content block in an earlier turn, but the result capture doesn't pick it up.

Post-processing logs confirm nothing was posted:
`No buffered inline comments`

## To Reproduce
1. Configure the `code-review@claude-code-plugins` plugin as shown in the workflow below.
2. Open a PR with non-trivial code changes (mine had ~1000 lines across migrations, API hooks, and components).
3. Add a label to trigger the workflow.
4. The action runs for ~7 minutes, completes successfully (exit code 0, 24 turns).
5. No comment appears on the PR.

## Expected behavior
The review summary should be posted as a PR comment. The plugin generates a thorough review (in my case it correctly identified 2 real bugs with file/line references), but the output is lost because the session result is empty.

## Workflow yml file
```yaml
name: Claude Code Review

on:
pull_request:
types: [opened, synchronize, ready_for_review, reopened, labeled]

jobs:
claude-review:
if: contains(github.event.pull_request.labels.*.name, 'claude-review')
runs-on: ubuntu-latest
permissions:
contents: read
pull-requests: write
issues: write
id-token: write

steps:
- name: Checkout repository
uses: actions/checkout@v4
with:
fetch-depth: 1

- name: Run Claude Code Review
id: claude-review
uses: anthropics/claude-code-action@v1
with:
claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
plugin_marketplaces: '[https://github.com/anthropics/claude-code.git](https://github.com/anthropics/claude-code.git)'
plugins: 'code-review@claude-code-plugins'
prompt: '/code-review:code-review ${{ github.repository }}/pull/${{ github.event.pull_request.number }}'
claude_args: "--allowedTools Bash(gh:*)"
show_full_output: true
```

Additional context

Key details from the session log with show_full_output:
true:

- Claude Code v2.1.79, models: claude-opus-4-6 (outer),
claude-sonnet-4-6 + claude-opus-4-6 (review agents)
- Session completed normally: "stop_reason": "end_turn",
24 turns, ~7 min
- The review text was output in message
msg_01A2MbGXySsa12kEoF6QZHZ4 as a text content block, but
that same message also contained a TodoWrite tool call.
The plugin's final message was the TodoWrite response,
leaving "result": "".
- The --allowedTools Bash(gh:*) flag was added to grant gh
CLI access, which resolved an earlier permission denial
issue — but the plugin never reaches the point of calling
gh to post comments because it ends on the TodoWrite
instead.

Start with the claude-code-action entry point that captures the session result and the code-review plugin invocation described in the workflow YAML. Reproduce the run with show_full_output enabled, then trace how text content is selected when the final turn also contains a TodoWrite call. Done means the generated review summary is captured and posted as a pull request comment.

Written by the indexing model from the issue text.