Skip to content
Agentmart
Submit a tool

requesting-code-review

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

Skill by transparent-pegasus

IndexedMIT★ 12
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/transparent-pegasus-requesting-code-review

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 26, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/requesting-code-review

What's inside

Skill name
requesting-code-review
Description
Use when completing tasks, implementing major features, or before merging to verify work meets requirements
License
MIT
Instructions
104 lines in SKILL.md · view on GitHub

Files (2)

  • SKILL.md
  • review-brief.md

SKILL.md (first lines)

# Requesting Code Review

**Placeholder resolution:** `<KEY>` placeholders in this file (such as `<BASE_BRANCH>`) resolve from the `Herdrpowers Configuration` section of the repository's `CLAUDE.md` / `AGENTS.md`. If that section is missing, initialize it with the pack's init workflow (`/herdrpowers:init` on Claude Code plugin installs; `commands/init.md` otherwise).

Delegate the review to a **herdr sibling pane via a reset-backed submit** to catch issues before they cascade. The reviewing pane starts from a reset session, so it sees the work product and never the implementer's thought process — that is where review independence comes from in this pack, not from a separate named agent or a different pane ID.

**Core principle:** Review early, review often. Do not review your own work inside an unreset session. A reset-backed delegation may reuse any idle pane of the resolved role's agent type(s), including the pane that wrote the code.

**Gate:** the whole-branch review this skill requests is the `assignments.final-branch-review` gate; per-task reviews are `assignments.task-review`. Both resolve from `.herdrpowers/config.yaml` over the pack's `orchestration/roles.yaml` defaults — see "Delegation tasks and their assignments" in `orchestration`. When the resolved role binds to a list of agent types, the review runs once per entry, in separate panes, and their findings merge into one list. A disabled gate means the review does not run and the final report says so. It never means an unreset implementing session reviews itself.

## When to Request Review

**Mandatory:**
- After each task in pane-driven development
- After completing a major feature
- Before merge to the base branch

**Optional but valuable:**
- When stuck (fresh perspective)
- Before refactoring (baseline check)
- After fixing a complex bug

## How to Request

**1. Record the range and build the diff package:**
```bash
BASE_SHA=$(git merge-base <BASE_BRANCH> HEAD)   # or the commit recorded before the task
HEAD_SHA=$(git rev-parse HEAD)
bash ../pane-driven-development/scripts/review-package "$PLAN_FILE" "$BASE_SHA" "$HEAD_SHA"   # prints the path
```
`PLAN_FILE` scopes the package to that plan's workspace. In plan-less flows (`/quick`, ad-hoc reviews) pass `-`.
The package never enters your own context — you pass the path, the pane reads the file.

**2. Pick the pane:**
Resolve the gate's `role` from the merged configuration. Pick any idle pane of that role's agent type(s). The pane that implemented is fine when the submit resets the session. If no pane of the needed type is idle but one exists and is busy, wait; do not interrupt. If none exists, fall back per `orchestration`, or state in the final report that the review could not be delegated.

**3. Delegate:**
Write the review contract from [review-brief.md](review-brief.md) to a file, then submit the one-line instruction through `using-herdr-sibling-panes` (`composer-submit.sh`) and wait for the completion marker. Never submit an agent-pane prompt with `pane run`.

**Placeholders:**
- `[DESCRIPTION]` - Brief summary of what you built
- `[PLAN_OR_REQUIREMENTS]` - What it should do
- `[BASE_SHA]` / `[HEAD_SHA]` - The range under review
- `[DIFF_FILE]` - The path `review-package` printed

**4. Act on feedback:**
- Read the review from its output file, not from pane scrollback
- Fix Critical issues immediately
- Fix Important issues before proceeding
- Note Minor issues for later
- Rule on every line of the reviewer's "Declined to judge" list: each set-aside behavior either stands, with your reason, or becomes a finding. None is dropped silently
- Push back if the reviewer is wrong (with reasoning) — see `receiving-code-review`

## Example

```
[Just completed Task 2: Add verification function]

You: Delegating the review before proceeding.

BASE_SHA=a7981ec   # recorded before Task 2's implementer was delegated
HEAD_SHA=3df7661
[review-package docs/plans/feature.md a7981ec 3df7661 → /repo/.herdrpowers/pdd/feature/review-a7981ec..3df7661.diff]
[Write the review contract to /repo/.herdrpowers/pdd/feature/task-2-review-brief.md]
[composer-submit.sh w2:p19 "<one-line instruction>"; pane wait-output REVIEW_OK_9C4A]

[Read /repo/.herdrpowers/pdd/feature/task-2-review.md]:
  Strengths: Clean architecture, tests assert the requirement
  Issues:
    Important: Missing progress indicators
    Minor: Magic number (100) for reporting interval
  Assessment: Ready to proceed

You: [Delegate one fix with both findings]
[Continue to Task 3]
```

## Common Rationalizations

| Excuse | Reality |
|--------|---------|
| "I'll just review the diff myself instead of delegating" | You're the orchestrator — reviewing the diff inline burns the context window you need to keep driving the work, and it is not an independent review. Delegate via a reset-backed submit: the diff and the evaluation live in that session, and only the findings come back to you. |
| "The reviewing pane needs my whole session history to understand the change" | Hand it precisely crafted context, never your session's history. That keeps the review on the work product, not your thought process. |
| "The only idle pane is the one that wrote the code" | Reuse it after a reset-backed submit. Independence is the fresh session, not a different pane ID. |

## Red Flags

**Never:**
- Skip review because "it's simple"
- Review the change inside an unreset implementing session
- Submit the review prompt with `pane run`, or without a completion marker
- Accept "looks good" without a written report file
- Ignore Critical issues
- Proceed with unfixed Important issues
- Argue with valid technical feedback

**If the reviewer is wrong:**
- Push back with technical reasoning
- Show code/tests that prove it works
- Request clarification

See the template at: [review-brief.md](review-brief.md)

From transparent-pegasus/herdrpowers, MIT.

More from transparent-pegasus/herdrpowers

Signals

Stars
12
Forks
0
Open issues
1
Last commit (this folder)
Oct 9, 2026 (19 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from transparent-pegasus

Cite this page

requesting-code-review skill. Agent Mart. https://agentmart.dev/skills/transparent-pegasus-requesting-code-review (last updated 2026-10-24)