sun-praise · Svtter · May 23, 2026 · May 23, 2026
diff --git a/.github/workflows/pi-multi-review.yml b/.github/workflows/pi-multi-review.yml
@@ -0,0 +1,34 @@
+name: Pi Multi-Review
+
+on:
+  pull_request:
+    types: [opened, synchronize, reopened, ready_for_review]
+  issue_comment:
+    types: [created]
+
+jobs:
+  multi-review:
+    # Skip draft PRs and cross-repo PRs; only trigger on /multi-review comments
+    if: >
+      (github.event_name == 'pull_request' && github.event.pull_request.draft == false && github.event.pull_request.head.repo.full_name == github.repository) ||
+      (github.event_name == 'issue_comment' && github.event.issue.pull_request != null && startsWith(github.event.comment.body, '/multi-review'))
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      pull-requests: write
+      issues: write
+    steps:
+      - name: Checkout PR head
+        uses: actions/checkout@v6
+        with:
+          repository: ${{ github.event.pull_request.head.repo.full_name || github.event.issue.pull_request.head.repo.full_name }}
+          ref: ${{ github.event.pull_request.head.ref || github.event.issue.pull_request.head.ref }}
+
+      - name: Run Pi Multi-Review
+        uses: ./pi-multi-review
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          provider: ${{ vars.PI_PROVIDER || 'anthropic' }}
+          model: ${{ vars.PI_MODEL || 'claude-sonnet-4-6' }}
+          api-key: ${{ secrets.PI_API_KEY }}
+          language: Chinese
diff --git a/README.md b/README.md
@@ -34,6 +34,7 @@ npx skills add sun-praise/opencode-actions
 - `architect-review`: architecture-level PR review focusing on coupling, layering, and structural concerns
 - `feature-missing`: audits PR implementation against linked issue spec to find missing features
 - `spec-coverage`: cross-references project spec/task files against PR implementation to find planned but unimplemented features
+- `pi-multi-review`: multi-agent parallel PR review using pi-coding-agent-action with pi-parallel-agents team mode
 - `github-run-opencode`: one-step wrapper for the common `opencode github run` workflow
 - `setup-opencode`: installs OpenCode, restores a dedicated cache, and exports the binary path
 - `run-opencode`: runs `opencode` with optional retry logic for flaky GitHub network failures
@@ -150,6 +151,7 @@ Unlike `feature-missing` (which checks PR self-described scope), `spec-coverage`
 | `architect-review` | PR diff + project conventions | Coupling, layering, module placement, structural concerns |
 | `feature-missing` | PR title/body + linked issues | PR self-described scope completeness |
 | `spec-coverage` | Project spec/task files | Full planned scope vs implementation |
+| `pi-multi-review` | PR diff (via pi-agent) | Multi-reviewer parallel review: quality, security, performance, architecture |
 
 ## setup-opencode
 
@@ -205,6 +207,7 @@ uses: sun-praise/opencode-actions/review@v2
 uses: sun-praise/opencode-actions/architect-review@v2
 uses: sun-praise/opencode-actions/feature-missing@v2
 uses: sun-praise/opencode-actions/spec-coverage@v2
+uses: sun-praise/opencode-actions/pi-multi-review@v2
 uses: sun-praise/opencode-actions/github-run-opencode@v2
 uses: sun-praise/opencode-actions/setup-opencode@v2
 uses: sun-praise/opencode-actions/run-opencode@v2

diff --git a/pi-multi-review/README.md b/pi-multi-review/README.md
@@ -0,0 +1,130 @@
+# Pi Multi-Review Action
+
+Multi-agent parallel PR review using [pi-coding-agent-action](https://github.com/shaftoe/pi-coding-agent-action) with [pi-parallel-agents](https://github.com/messense/pi-parallel-agents) team mode.
+
+Multiple reviewer agents (quality, security, performance, architecture) review your PR simultaneously, then a synthesizer agent produces a single consolidated review comment.
+
+## Usage
+
+```yaml
+name: Pi Multi-Review
+
+on:
+  pull_request:
+    types: [opened, synchronize, reopened, ready_for_review]
+  issue_comment:
+    types: [created]
+
+jobs:
+  multi-review:
+    if: >
+      (github.event_name == 'pull_request' && github.event.pull_request.draft == false) ||
+      (github.event_name == 'issue_comment' && github.event.issue.pull_request != null && startsWith(github.event.comment.body, '/multi-review'))
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      pull-requests: write
+      issues: write
+    steps:
+      - uses: actions/checkout@v6
+
+      - uses: sun-praise/opencode-actions/pi-multi-review@v2
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          provider: anthropic
+          model: claude-sonnet-4-6
+          api-key: ${{ secrets.PI_API_KEY }}
+          language: Chinese
+```
+
+## Inputs
+
+| Input | Required | Default | Description |
+|-------|----------|---------|-------------|
+| `github-token` | Yes | - | GitHub token for API access |
+| `provider` | Yes | - | LLM provider (openai, anthropic, google, etc.) |
+| `model` | No | `claude-sonnet-4-6` | Default model for all reviewers |
+| `api-key` | No | - | LLM provider API key |
+| `reviewers-config` | No | `""` | Inline YAML to override default reviewer personas |
+| `max-tokens` | No | `4096` | Max tokens per reviewer response |
+| `diff-max-lines` | No | `2000` | Max diff lines included in review |
+| `diff-ignore-patterns` | No | `""` | File patterns to exclude from diffs |
+| `base-url` | No | `""` | Provider base URL override (for proxies) |
+| `thinking-level` | No | `medium` | Model thinking level (off, low, medium, high) |
+| `language` | No | `Chinese` | Output language for reviews |
+
+## Default Reviewers
+
+When no `reviewers-config` is provided, 4 built-in reviewers run in parallel:
+
+| Reviewer | Focus |
+|----------|-------|
+| **quality** | Readability, naming, error handling, DRY, dead code |
+| **security** | Input validation, injection, OWASP Top 10, secrets exposure |
+| **performance** | Algorithmic complexity, N+1 queries, memory, caching |
+| **architecture** | Coupling, separation of concerns, API design, scalability |
+
+A **synthesizer** agent merges all findings into one structured PR comment with:
+- Merge decision (可合并 / 有条件合并 / 不可合并)
+- Critical issues, warnings, suggestions
+- Cross-validated findings (flagged by multiple reviewers)
+
+## Custom Reviewers
+
+Provide your own reviewers via `reviewers-config`:
+
+```yaml
+- uses: sun-praise/opencode-actions/pi-multi-review@v2
+  with:
+    github-token: ${{ secrets.GITHUB_TOKEN }}
+    provider: anthropic
+    api-key: ${{ secrets.PI_API_KEY }}
+    reviewers-config: |
+      reviewers:
+        - name: security
+          role: Security Expert
+          prompt: |
+            Focus only on security vulnerabilities in this diff:
+            {{diff}}
+
+            Report critical, warning, and info findings.
+        - name: testing
+          role: Test Coverage Analyst
+          prompt: |
+            Check test coverage for changes in this diff:
+            {{diff}}
+
+            Are new code paths adequately tested?
+      synthesizer:
+        name: synthesizer
+        role: Review Synthesizer
+        prompt: |
+          Combine these findings into a structured review:
+          {task:security-review}
+          {task:testing-review}
+          Output as markdown.
+```
+
+## Template Variables
+
+Reviewer prompts support these placeholders:
+
+| Variable | Description |
+|----------|-------------|
+| `{{diff}}` | Full PR diff |
+| `{{title}}` | PR title |
+| `{{body}}` | PR description |
+| `{{files}}` | List of changed files |
+| `{{language}}` | Configured output language |
+
+Synthesizer prompts can reference reviewer outputs with `{task:<name>-review}`.
+
+## Requirements
+
+- Linux runner (GitHub Actions `ubuntu-latest`)
+- Python 3 with PyYAML (pre-installed on GitHub-hosted runners)
+- A valid LLM API key for the configured provider
+
+## Security Note
+
+This action delegates to `shaftoe/pi-coding-agent-action@v2`. For production use, consider pinning to a specific commit SHA instead of the `v2` tag to protect against supply-chain attacks. See the [GitHub Actions security guide](https://docs.github.com/en/actions/security-for-github-actions/security-guides/security-hardening-for-github-actions#using-third-party-actions).
diff --git a/pi-multi-review/action.yml b/pi-multi-review/action.yml
@@ -0,0 +1,93 @@
+name: Pi Multi-Review
+description: Multi-agent parallel PR review using pi-coding-agent-action with pi-parallel-agents team mode.
+
+inputs:
+  github-token:
+    description: GitHub token for API access.
+    required: true
+  provider:
+    description: LLM provider (e.g. openai, anthropic, google).
+    required: true
+  model:
+    description: Default model for all reviewers (e.g. claude-sonnet-4-6, gpt-4o).
+    required: false
+    default: "claude-sonnet-4-6"
+  api-key:
+    description: API key for the LLM provider. Can also be set via provider-specific env vars.
+    required: false
+    default: ""
+  reviewers-config:
+    description: >-
+      Inline YAML string defining custom reviewer personas. When empty, uses the 4 built-in
+      reviewers (quality, security, performance, architecture). Each reviewer has: name, role,
+      prompt, model (optional).
+    required: false
+    default: ""
+  max-tokens:
+    description: Maximum tokens per reviewer response.
+    required: false
+    default: "4096"
+  diff-max-lines:
+    description: Maximum number of diff lines to include in PR diff.
+    required: false
+    default: "2000"
+  diff-ignore-patterns:
+    description: Space-separated file patterns to exclude from PR diffs.
+    required: false
+    default: ""
+  base-url:
+    description: Optional override for the provider base URL (e.g. for proxies or compatible gateways).
+    required: false
+    default: ""
+  thinking-level:
+    description: Thinking level for the model (off, low, medium, high).
+    required: false
+    default: "medium"
+  language:
+    description: Language for review output (e.g. Chinese, English).
+    required: false
+    default: "Chinese"
+
+runs:
+  using: composite
+  steps:
+    - if: ${{ runner.os != 'Linux' }}
+      shell: bash
+      run: |
+        set -euo pipefail
+        printf 'pi-multi-review currently supports Linux runners only\n' >&2
+        exit 1
+
+    - name: Build multi-review prompt
+      id: build-prompt
+      shell: bash
+      env:
+        INPUT_REVIEWERS_CONFIG: ${{ inputs.reviewers-config }}
+        INPUT_MODEL: ${{ inputs.model }}
+        INPUT_LANGUAGE: ${{ inputs.language }}
+        INPUT_MAX_TOKENS: ${{ inputs.max-tokens }}
+        ACTION_PATH: ${{ github.action_path }}
+      run: |
+        set -euo pipefail
+        python3 "$ACTION_PATH/scripts/build-prompt.py" \
+          --reviewers-config "$INPUT_REVIEWERS_CONFIG" \
+          --default-model "$INPUT_MODEL" \
+          --language "$INPUT_LANGUAGE" \
+          --max-tokens "$INPUT_MAX_TOKENS" \
+          --action-path "$ACTION_PATH" \
+          --output "$RUNNER_TEMP/pi-multi-review-prompt.txt"
+        echo "prompt-file=$RUNNER_TEMP/pi-multi-review-prompt.txt" >> "$GITHUB_OUTPUT"
+
+    - name: Run pi multi-review
+      uses: shaftoe/pi-coding-agent-action@v2
+      with:
+        github_token: ${{ inputs.github-token }}
+        provider: ${{ inputs.provider }}
+        model: ${{ inputs.model }}
+        token: ${{ inputs.api-key }}
+        prompt: ${{ steps.build-prompt.outputs.prompt-file }}
+        extensions: "npm:pi-parallel-agents"
+        thinking_level: ${{ inputs.thinking-level }}
+        diff_max_lines: ${{ inputs.diff-max-lines }}
+        diff_ignore_patterns: ${{ inputs.diff-ignore-patterns }}
+        base_url: ${{ inputs.base-url }}