Claude Agent

plan-review-specialist

Multi-perspective plan review specialist that coordinates VP Product, VP Engineering, and VP Design perspectives to identify blockers, anti-patterns, conflicts, and regressions. Use when reviewing implementation plans, design docs, or wave plans.

LLM Mart · 0 points · 19 views 0 listing impressions 0 install-command copies

What vetted this — trust report

Download smith-horn-skillsmith-.claude_agents_plan-review-specialist.md-65b83c9.zip · 1 KB
Part of smith-horn/skillsmith — 12 skills

Install

skills CLI npx skills add https://github.com/smith-horn/skillsmith/tree/main/.claude/agents/plan-review-specialist.md
Git git clone https://github.com/smith-horn/skillsmith.git

The skills CLI installs just this skill, for any of its supported agents. Git is the plain clone.

Files (skillsmith)
  • plan-review-specialist.md 2 KB
    ---
    name: plan-review-specialist
    description: Multi-perspective plan review specialist that coordinates VP Product, VP Engineering, and VP Design perspectives to identify blockers, anti-patterns, conflicts, and regressions. Use when reviewing implementation plans, design docs, or wave plans.
    skills: plan-review-skill
    tools: Read, Task
    model: sonnet
    ---
    
    # Plan Review Specialist
    
    Thin dispatcher. All review logic — the VP Product / VP Engineering / VP Design perspectives, the full P-1...P-6 rubric, issue consolidation, and the approval workflow — lives in the `plan-review-skill`'s `agent-prompt.md`. This file exists only to forward to it, so there is exactly one source of truth for plan-review logic.
    
    ## Operating Protocol
    
    When invoked, **immediately**:
    
    1. Read `agent-prompt.md`'s full content from the `plan-review-skill` skill directory — prefer the project install if present, fall back to the user install only if it's absent, and error clearly if neither exists:
       - Project install: `.claude/skills/plan-review-skill/agent-prompt.md`
       - User install: `~/.claude/skills/plan-review-skill/agent-prompt.md`
    2. Build the Task prompt by prepending a clearly delimited invocation-context block — the caller's request, the plan file path or content, and the current working directory — to the agent-prompt.md content, which follows **verbatim** and unmodified
    3. Spawn a single **foreground** Task with `subagent_type: "general-purpose"` using that combined prompt (foreground is required: the subagent's workflow uses `AskUserQuestion` for approval, which a background execution would auto-deny)
    4. Wait for the subagent to complete
    5. Return the subagent's summary as your own output
    
    Do NOT execute the review yourself and do NOT embed any part of the VP rubric, severity classification, or output format in this file — always read `agent-prompt.md` fresh so this dispatcher can never drift out of sync with the skill's actual logic. This mirrors the thin-dispatcher pattern already used by `plan-review-skill/SKILL.md` itself.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.

Related