| name | plan-review |
| description | Use when explicitly asked to walk through a plan or list of issues one-by-one, with the user deciding each via AskUserQuestion. Triggers only on a request to review/triage a plan interactively — NOT on any casual mention of a feature, fix, or bug. |
Plan Review
Review and discuss issues that need fixing or implementing, one at a time.
Process
-
Identify issues to review from conversation context. Look at the current conversation for a plan, a list of issues, or any prior discussion of features, fixes, or bugs. Work from whatever is already in context — do NOT scan the codebase independently for issues.
-
For each issue, explain it clearly. Describe the issue as though the user doesn't know much about the code or the problem. Cover:
- What the issue is and where it lives in the codebase
- Why it matters (impact, risk, or benefit of fixing it)
- What the possible fix or implementation options are, with trade-offs
-
Use AskUserQuestion for each issue. Present the user with clear options for how to proceed — e.g., fix now, skip, defer, or take an alternative approach. Let the user drive the decision.
-
Ask clarifying questions freely. If anything is unclear at any point — about the codebase, the user's intent, or the best approach — use AskUserQuestion immediately rather than guessing.
-
Proceed through all issues sequentially. Don't rush or batch. One issue at a time, fully resolved before moving to the next.
-
After all issues are decided, summarize, record, and delegate. Recap the decisions made, then record the chosen actions into the task list with TaskCreate — capturing any sequential (→) or parallel (||) shape between them. Then DELEGATE the chosen fixes to sub-agents per the orchestrator rules in ~/.claude/CLAUDE.md. This skill itself does NOT start writing code — it routes the decided work to sub-agents.
