메뉴
Rules for AI sessions that address PR review comments. Covers fetching review comments, verifying findings, making fixes, and pushing changes. Read this at the start of every review-addressing session.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Instructions for contextual knowledge retrieval, tagging, search, and rating. Referenced by plan-session, implementation-session, and review-pr-comments-session skills.
Rules for AI implementation sessions in a wade-managed worktree. Covers worktree safety, commit conventions, syncing with main, PR summaries, and session closing. Read this at the start of every work session.
Rules for AI planning sessions launched by `wade plan`. Covers plan file authoring, complexity tagging, and session boundaries. Read this at the start of every planning session.
Create one or more GitHub issues via wade — for a single bug/feature OR a full plan/PRD breakdown. Assesses scope, proposes single or multi-issue breakdown with reasoning, gets user confirmation, writes plan files, creates issues via wade task create, and informs the user of next steps. ALL steps are mandatory — do not stop after planning. Use whenever the user asks to create a GitHub issue, regardless of scope.
Analyze a set of GitHub issues and identify dependency relationships between them. Reads issue titles and bodies, determines execution order, and outputs a structured dependency file that wade uses to generate Mermaid diagrams and update issue bodies. Use when multiple related issues need ordering.
| name | review-pr-comments-session |
| description | Rules for AI sessions that address PR review comments. Covers fetching review comments, verifying findings, making fixes, and pushing changes. Read this at the start of every review-addressing session. |
These rules govern AI sessions that address PR review comments. Read and follow them before performing any other action.
Run wade and gh commands with the required permissions/capabilities (not
in sandboxed mode). Do not "try sandbox first" — run them unsandboxed from the
start.
Always inform the user before running wade commands, reviews, or
session lifecycle operations. Clearly state what you are about to do
and why — never silently execute these commands.
When starting a workflow step, announce it: "I'm now syncing your branch with main..."
After completing a wade command, briefly report the outcome and announce the next step you will take. The next step depends on where you are in the workflow — for example:
"Sync complete — your branch is up to date with main. Now running wade review-pr-comments-session done..."
"Push complete — PR updated. Now presenting the workflow recap and suggesting you exit..."
{user_interaction_prompt}
gh issue createNEVER use gh issue create or the GitHub API to create issues directly.
Always use wade task create for interactive issue creation.
Read @.claude/skills/knowledge/SKILL.md for knowledge operations (search, tagging, rating, adding entries).
Before addressing review comments, search for knowledge relevant to the files and topics being reviewed (do not dump all entries). Past gotchas about the files you're editing matter most here. Rating is required for each entry you open and evaluate — not per search call. See the decision tree in @.claude/skills/knowledge/SKILL.md for how to decide between up, down, or no rating.
Run wade review-pr-comments-session check as your first action:
IN_WORKTREE — you may proceed with work (code changes, commits, etc.)IN_MAIN_CHECKOUT — editing any source file is forbidden. Tell the
human to run wade review pr-comments <issue> from the main checkout.NOT_IN_GIT_REPO — you are not inside a git repository.Use wade review-pr-comments-session fetch <issue-number> to fetch all unresolved PR review
comments. This outputs formatted markdown with:
Run this command first to understand what needs to be addressed.
Always verify each finding against the current code before fixing it. Automated review tools (CodeRabbit, etc.) can be wrong — they may flag code that is actually correct, or suggest changes that don't apply.
CodeRabbit comments include a 🤖 Prompt for AI Agents section — this is
the primary instruction to follow. The full comment body provides additional
context (rationale, proposed diffs, etc.) but the AI-agent prompt is the
actionable instruction.
Severity indicators:
Human reviewer comments use the full comment body as the instruction. Follow reviewer intent — ask clarifying questions in your commit message if the request is ambiguous.
Address comments file-by-file or in logical groups. Each commit should be cohesive — don't mix unrelated fixes.
After addressing a review comment, resolve the corresponding thread on GitHub:
wade review-pr-comments-session resolve <thread-node-id>
The thread ID is included in the output of wade review-pr-comments-session fetch.
Use Conventional Commits format:
fix: address review comment — <brief description>fix: address review comments (for multiple related fixes)refactor: address review feedback on <component>Run the project's test suite after making changes to ensure nothing is broken. Add tests if a review comment identified a missing test case.
Reference section — this describes how syncing works and how to handle conflicts. The actual sync is performed as part of the closing workflow below (Step 2). Do not run sync separately.
Exit code 0 — Success: Branch is up to date with main. Proceed to closing.
Exit code 2 — Conflict: The merge is paused due to conflicts:
git diff --name-only --diff-filter=U to list conflicted filesgit add <file1> <file2> ...git commit --no-editwade review-pr-comments-session sync --json to verify cleanExit code 4 — Pre-flight failure: Report the issue and suggest how to fix it.
Before closing the session, write PR-SUMMARY.md in the worktree root.
wade review-pr-comments-session done reads this file to update the PR body.
Never commit this file — it is a session artifact (already in
.gitignore).
## What was addressed
[Summary of review comments handled]
## Changes
- Fixed X per reviewer feedback
- Resolved CodeRabbit finding on Y
## Remaining
[Any threads intentionally left unresolved, with reasoning]
NEVER create Pull Requests manually (gh pr create) or push branches
directly.
To finalize your work, follow these steps in order:
Step 1 — Write PR summary:
Write PR-SUMMARY.md in the worktree root (see format above). If the file
already exists, update it.
Step 2 — Sync with main:
wade review-pr-comments-session sync --json
Step 3 — Done:
wade review-pr-comments-session done
wade review-pr-comments-session done pushes changes to the existing PR branch.
This is a mandatory completion step. If it fails, debug and fix the error — do NOT bypass it.
Step 4 — Present results to the user:
Provide a brief workflow recap and current state summary, then suggest the user exits the session.
Workflow recap (list only the steps you actually performed):
wade review-pr-comments-session fetch)wade review-pr-comments-session sync)wade review-pr-comments-session done)Current state:
What happens next:
Use your tool's native question component to ask: "Want any further changes, or is the session complete?" Apply them and repeat Steps 1–4 if so. Otherwise, suggest the user exits so wade can continue the workflow.
At the start of this session, use your tool's native task/todo tracking mechanism to populate a checklist with the workflow steps below. This ensures you complete every mandatory step and the user can track progress.
wade review-pr-comments-session checkwade review-pr-comments-session fetch)wade knowledge get --search <topic> or wade knowledge get --tag <tag>)up/down when appropriate (wade knowledge rate <id> up/down per the decision tree in knowledge skill); otherwise intentionally leave them unratedwade review-pr-comments-session sync --json)wade review-pr-comments-session done)