@agentoom/pi-reviewer

Post-task multi-area code review extension for pi by agentoom.com.

After every task, pi-reviewer prompts you to run a structured critique across 7 specialised areas — or you can call /review anytime.

v1.7.0 — You can now configure a dedicated model for reviews (global via /review-set-model, per-project via /review-model-project, or per-review via the picker flow). See Review Model Configuration below.

Install

pi install npm:@agentoom/pi-reviewer

Usage

Commands

| Command | Description | |---------|-------------| | /review | Open the review area picker and run selected reviews | | /review-toggle | Toggle the auto‑offer prompt on/after each task | | /review-disable-project | Disable auto-review for the current project (writes .pi/review-config.json) | | /review-enable-project | Re-enable auto-review for the current project | | /review-set-model | Set the model used for code reviews globally (persists across sessions) | | /review-model-project | Set the model used for code reviews in this project (writes .pi/review-config.json) |

Review Model Configuration

By default, reviews use whatever model is currently active in your pi session. You can override this at three levels:

1. Per-review override (highest priority): When running a review (via /review or the auto‑offer prompt), you'll see a "Review model?" step that shows the currently effective model. Choose "Choose a different model for this review" to pick a one‑off model for this review only. The override is cleared after the review completes.

2. Project-level config: Run /review-model-project to set a default review model for the current project. This writes a reviewModel field to .pi/review-config.json. Run the command again and select "None" to clear it. Example config:

{
  "autoReview": true,
  "reviewModel": "anthropic/claude-sonnet-4-20250514"
}

3. Global config: Run /review-set-model to set a review model globally — it persists across all sessions and projects (stored in the pi session state). Run the command again and select "None" to clear it.

Resolution priority: Per‑review override → Project config → Global config → Current session model (fallback).

Review Scope

Before each review, you're asked to choose the scope:

• Last prompt only (default) — only the most recent user prompt and the agent's response are reviewed. This is the recommended default and avoids reviewing irrelevant context from earlier in the session.
• Entire session — the full conversation from the session (or from the last compaction point) is reviewed.

Apply Review Findings

After the review completes, the results appear directly in the chat transcript as a styled 📋 Code Review Report entry. These entries are rendered in the TUI but do not participate in LLM context — subsequent prompts won't see or be confused by them.

pi-reviewer scans the results for actionable issues (sections marked with ❌ or ⚠️ containing bullet points). If issues are found, you'll be prompted:

• Apply fixes now — you'll be asked for optional instructions (e.g. "skip the scope finding about utils.ts, it's a necessary dependency"). The LLM reads the review findings, follows your guidance, edits the affected files, and reports what was changed. Fix results appear as a 🔧 Fixes Applied entry in the chat.
• Skip — the review report stays visible but no fixes are applied.

If the review found no issues, the prompt is skipped.

Auto‑offer

By default, pi-reviewer prompts you after every task completes. Choose:

• Review (all 7 areas) — you'll be asked about review scope and model, then all areas run
• Pick which areas to review — you'll be asked about review scope and model, then the checkbox picker opens
• Disable auto-offer — switch to manual mode permanently
• Skip — dismiss for this task only

When auto‑offer is disabled, 🔍 auto-review disappears from the footer, and pi-reviewer stays silent until you run /review.

Toggle it back on with /review-toggle.

Disabling Auto-Review

For the current project (persistent)

Run /review-disable-project to write autoReview: false to .pi/review-config.json. The extension will skip auto-offering after each task. Use /review-enable-project to undo.

You can also create .pi/review-config.json manually:

{
  "autoReview": false,
  "reviewModel": "anthropic/claude-sonnet-4-20250514"
}

For the current session only

Set an environment variable before starting pi:

REVIEWER_DISABLE_SESSION=true pi

This suppresses the auto-review prompt for that session only — other pi sessions (without the env var) are unaffected. The env var takes priority over the project config.

Review Areas

| Area | Question | |------|----------| | General | Did the implementation satisfy the original task? | | Security | Did it introduce vulnerabilities? | | Code Quality | Is it maintainable and consistent? | | UI/UX | Is the user-facing implementation complete? | | Testing | Is it adequately tested? | | Performance | Could it cause performance problems? | | Scope | Did the agent change more than it should have?Distinguishes truly out-of-scope changes from functional dependencies (supporting code that was required for the feature). |

Checkbox Picker

• ↑↓ — navigate
• Space — toggle an area
• a — toggle all areas
• Enter — run selected reviews
• Esc — cancel

Development

pi-reviewer/
├── package.json
├── index.ts          # Extension entry point
└── README.md

Links

• agentoom.com
• npm: @agentoom/pi-reviewer

Run against a local path:

pi install ./pi-reviewer