This page documents the requesting-code-review skill and the code-reviewer subagent it dispatches. It covers when to invoke the review process, how git SHAs are used to scope the diff, issue severity classification, and how review integrates into the larger development workflows.
For the broader workflow that calls into code review, see Subagent-Driven Development and Executing Plans in Batches. For guidance on receiving and responding to review feedback from the agent perspective, see Other Essential Skills.
The requesting-code-review skill (skills/requesting-code-review/SKILL.md) provides a standardized protocol for dispatching a fresh code-reviewer subagent against a specific range of commits. Its core principle is review early, review often — catching issues task-by-task rather than accumulating them until merge time.
The skill is invoked by the coordinating agent, not by the implementer. It is the third stage of the per-task review sequence in Subagent-Driven Development, following the implementer and spec-compliance reviewer stages.
Code review has mandatory and optional invocation contexts:
| Context | Trigger | Source |
|---|---|---|
| Subagent-Driven Development | After each task completes spec compliance review | skills/requesting-code-review/SKILL.md15-16 |
| Executing Plans | After each batch of tasks (default: 3 tasks) | skills/requesting-code-review/SKILL.md84-86 |
Before merge to main | Mandatory | skills/requesting-code-review/SKILL.md17 |
| When stuck | Optional — fresh perspective | skills/requesting-code-review/SKILL.md20-21 |
| Before refactoring | Optional — baseline check | skills/requesting-code-review/SKILL.md22 |
| After fixing complex bug | Optional | skills/requesting-code-review/SKILL.md23 |
Never skip because "it's simple" (skills/requesting-code-review/SKILL.md94).
Sources: skills/requesting-code-review/SKILL.md14-23 skills/subagent-driven-development/code-quality-reviewer-prompt.md1-7
The reviewing agent computes two SHAs that define the exact diff to review:
In practice, BASE_SHA is the commit before the task began, and HEAD_SHA is the commit after the implementer finished. This scopes the reviewer to only the changes made in the current task.
Sources: skills/requesting-code-review/SKILL.md27-29
superpowers:code-reviewerThe coordinating agent uses the Task tool with type superpowers:code-reviewer, filling in the template located at requesting-code-review/code-reviewer.md.
Required template placeholders:
| Placeholder | Content |
|---|---|
{WHAT_WAS_IMPLEMENTED} | Description of what was just built |
{PLAN_OR_REQUIREMENTS} | The task spec or requirements document |
{BASE_SHA} | Starting commit (before task) |
{HEAD_SHA} | Ending commit (after task) |
{DESCRIPTION} | Brief one-line summary |
Sources: skills/requesting-code-review/SKILL.md36-47 skills/subagent-driven-development/code-quality-reviewer-prompt.md10-17
The coordinator applies feedback according to severity classification before proceeding.
The code-reviewer subagent classifies every issue it finds into one of three tiers:
| Severity | Required Action | Effect on Progress |
|---|---|---|
| Critical | Fix immediately | Block all forward progress |
| Important | Fix before proceeding to next task | Block next task, but not review completion |
| Minor | Note for later | Non-blocking; log for backlog |
The coordinator may push back on any issue if the feedback is incorrect, but must provide technical reasoning and code/test evidence (skills/requesting-code-review/SKILL.md99-102).
Sources: skills/requesting-code-review/SKILL.md43-47 skills/requesting-code-review/SKILL.md94-102
Within subagent-driven development, code review is the third stage of a per-task pipeline:
Per-Task Pipeline in Subagent-Driven Development
Sources: skills/requesting-code-review/SKILL.md79-82 skills/subagent-driven-development/code-quality-reviewer-prompt.md1-19 skills/subagent-driven-development/spec-reviewer-prompt.md1-10
When using the executing-plans skill, code review happens at batch boundaries rather than per-task:
Batch Execution with Review Checkpoints
In this mode, BASE_SHA is the commit at the start of the batch, and HEAD_SHA is the commit at the end of the batch.
Sources: skills/requesting-code-review/SKILL.md83-87
Skill dispatch flow mapped to files:
| File | Role |
|---|---|
skills/requesting-code-review/SKILL.md | Master skill definition; protocol, when-to-use, SHA computation, issue handling |
skills/requesting-code-review/code-reviewer.md | Prompt template filled with placeholders before dispatch |
skills/subagent-driven-development/code-quality-reviewer-prompt.md | Coordinator-side instructions for dispatching the code-reviewer within the subagent workflow |
skills/subagent-driven-development/spec-reviewer-prompt.md | Stage 2 reviewer that gates access to code review |
skills/subagent-driven-development/implementer-prompt.md | Stage 1 implementer whose output feeds into both reviewers |
Sources: skills/requesting-code-review/SKILL.md1-106 skills/subagent-driven-development/code-quality-reviewer-prompt.md1-19 skills/subagent-driven-development/spec-reviewer-prompt.md1-62 skills/subagent-driven-development/implementer-prompt.md1-79
The skill explicitly prohibits several shortcuts (skills/requesting-code-review/SKILL.md93-102):
| Anti-pattern | Why It's Prohibited |
|---|---|
| Skip review because "it's simple" | Simple code has bugs too; omission compounds into later tasks |
| Ignore Critical issues and proceed | Critical issues block all forward progress by definition |
| Proceed with unfixed Important issues | Important issues must be resolved before the next task begins |
| Argue with valid feedback | Valid technical feedback must be applied, not dismissed |
If the reviewer is factually wrong, the correct response is to push back with: specific code references, test results that demonstrate correctness, and a request for clarification on the reviewer's concern.
Sources: skills/requesting-code-review/SKILL.md93-102
The following is a condensed trace of the review process from the skill's own example (skills/requesting-code-review/SKILL.md49-74):
[Task 2 just completed: Add verification function]
1. Coordinator computes SHAs:
BASE_SHA = a7981ec (last commit from Task 1)
HEAD_SHA = 3df7661 (current HEAD after Task 2)
2. Coordinator dispatches superpowers:code-reviewer:
WHAT_WAS_IMPLEMENTED: "Verification and repair functions for conversation index"
PLAN_OR_REQUIREMENTS: "Task 2 from docs/plans/deployment-plan.md"
BASE_SHA: a7981ec
HEAD_SHA: 3df7661
DESCRIPTION: "Added verifyIndex() and repairIndex() with 4 issue types"
3. Reviewer returns:
Strengths: Clean architecture, real tests
Issues:
Important: Missing progress indicators
Minor: Magic number (100) for reporting interval
Assessment: Ready to proceed (after fixing Important issue)
4. Coordinator fixes progress indicators, re-commits.
5. Coordinator continues to Task 3.
Refresh this wiki
This wiki was recently refreshed. Please wait 3 days to refresh again.