Requesting Code Review

Dispatch code-reviewer sub-agent using the task tool to catch issues before they cascade. The reviewer gets precisely crafted context for evaluation — never your session's history. This keeps the reviewer focused on the work product, not your thought process, and preserves your own context for continued work.

Core principle: Review early, review often.

When to Request Review

Mandatory:

• After each task in subagent-driven development
• After completing major feature
• Before merge to main

Optional but valuable:

• When stuck (fresh perspective)
• Before refactoring (baseline check)
• After fixing complex bug

How to Request

1. Get git SHAs:

BASE_SHA=$(git rev-parse HEAD~1)  # or origin/main
HEAD_SHA=$(git rev-parse HEAD)

2. Dispatch code-reviewer or code-quality-reviewer global agent using the task tool:

The code-reviewer and code-quality-reviewer agents are registered globally with appropriate review checklists. Invoke with context about what was implemented, plan reference, and git diff range. The agent will use git diff {BASESHA}..{HEADSHA} to inspect changes.

Context to include in your Task tool prompt:

• What you just built (brief summary)
• What it should do (plan or requirements reference)
• Starting commit SHA (BASE_SHA)
• Ending commit SHA (HEAD_SHA)

3. Act on feedback:

• Fix Critical issues immediately
• Fix Important issues before proceeding
• Note Minor issues for later
• Push back if reviewer is wrong (with reasoning)

Example

[Just completed Task 2: Add verification function]

You: Let me request code review before proceeding.

BASE_SHA=$(git log --oneline | grep "Task 1" | head -1 | awk '{print $1}')
HEAD_SHA=$(git rev-parse HEAD)

[Dispatch code-reviewer sub-agent using the `task` tool]
  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

[Sub-agent returns]:
  Strengths: Clean architecture, real tests
  Issues:
    Important: Missing progress indicators
    Minor: Magic number (100) for reporting interval
  Assessment: Ready to proceed

You: [Fix progress indicators]
[Continue to Task 3]

Integration with Workflows

Subagent-Driven Development:

• Review after EACH task
• Catch issues before they compound
• Fix before moving to next task

Executing Plans:

• Review after each batch (3 tasks)
• Get feedback, apply, continue

Ad-Hoc Development:

• Review before merge
• Review when stuck

Choosing the Reviewer

Order matters. Running quality review before spec compliance wastes effort polishing code that may not be what was asked for. See subagent-driven-development.

Writing the Dispatch Prompt

The reviewer starts with no context. What you write is everything they know — never your session history. A reviewer given a vague prompt returns vague findings.

The prompt contains, in this order:

1. What was built — two sentences
2. What it should do — the requirement or plan task text, pasted in full, not a file reference
3. The diff range — BASESHA and HEADSHA
4. Where to look — the files that changed
5. What you already know — deviations you made, and why
6. What you want back — the report format

Paste the requirement text. Telling the reviewer to read docs/plans/x.md costs a tool call and risks them reading the wrong task. You already have the text.

Do not tell them what you think. "I think the error handling is solid" biases the review toward agreement. State facts, ask for findings.

Review the implementation of Task 2 from the auth plan.

REQUIREMENT (verbatim from the plan):
  Add verifyIndex() and repairIndex() to src/index/verify.ts. verifyIndex()
  returns a list of issues across 4 types: orphaned, stale, duplicate,
  missing. repairIndex() fixes each type. Report progress every 100 items.

DIFF: git diff a7981ec..3df7661
FILES: src/index/verify.ts, src/index/types.ts, tests/index/verify.test.ts

KNOWN DEVIATION: used a WeakMap rather than the Map the plan specified,
so entries are collected with their request objects. Behavior identical.

Return: PASS or FAIL with file:line references for each finding.

Acting on Feedback

REQUIRED SUB-SKILL: Use receiving-code-review before implementing any suggestion. Reviewers are sometimes wrong, and blind implementation of a wrong suggestion is worse than no review.

Then re-review. A reviewer that found issues has not approved the fixes. Re-dispatch with the same context plus what changed. Skipping the re-review is how "fixed" issues survive to merge.

Red Flags

Never:

• Skip review because "it's simple"
• Ignore Critical issues
• Proceed with unfixed Important issues
• Argue with valid technical feedback
• Make the reviewer read the plan file instead of pasting the task text
• Include your own assessment in the prompt (biases the review)
• Accept a fix as done without re-review
• Review uncommitted work — commit first so the diff range is real

If reviewer wrong:

• Push back with technical reasoning
• Show code/tests that prove it works
• Request clarification

Common Mistakes

Dispatching with session context. The reviewer inherits your assumptions and reviews your reasoning rather than the code.

No diff range. The reviewer reads the whole file and comments on pre-existing code that is not part of this change.

Reviewing before committing. No stable range to diff, and the reviewer may see a half-saved state.

One reviewer for both spec and quality. They are different questions and merging them means the scope question gets less attention than it needs.

Treating the review as a formality. If every review returns "approved, no issues", either the prompt is biasing it or the wrong reviewer is being used.