pm%3areview（pm%3 查看）

Compatibility: If AskUserQuestion is unavailable, present options as a numbered list and wait for the user's reply. If Task is unavailable, run parallel steps sequentially.

Review a product roadmap for quality, prioritization, and strategic coherence.

Step 0: Parse arguments and load artifact

Accept a roadmap file path or slug. Default to the most recently modified file in .codevoyant/roadmaps/.

ROADMAP_PATH="${1:-}"
SILENT=false
[[ "$*" =~ --silent ]] && SILENT=true

If ROADMAP_PATH not provided, find the most recent file:

ls -t .codevoyant/roadmaps/*.md 2>/dev/null | head -1

Set ROADMAP_FILE. Read the file. Parse flags: --silent.

Step 1: Run parallel review agents

Launch 3 review agents (model: claude-haiku-4-5-20251001, run_in_background: true) in a single message:

• Agent R1 — Prioritization: Does the ordering across capability tiers reflect the stated strategic goal? Are any must-have outcomes deferred to Tier 2/3?
• Agent R2 — Capability quality: For each capability in each tier, check:

- Name is outcome-oriented (not "ship X" or "build Y") — if solution-phrased, flag CRITICAL - "What it enables" describes a user or business outcome, not a deliverable — if deliverable-phrased, flag CRITICAL - "Why now" rationale is present and specific — if absent, flag INFORMATIONAL - Key bets are present (2–4 bullets) — if missing, flag INFORMATIONAL

• Agent R3 — Strategic coherence: Does the roadmap tell a coherent story? Are themes consistent across tiers? Does the "What We Are NOT Doing" section contain explicit deferrals?

Wait for all three. Synthesize results.

Step 2: Two-pass classification

Pass 1 — CRITICAL (must resolve before approving)

• Capability name or "what it enables" phrased as a deliverable (ship/build/add/create)
• Missing strategic goal section
• Tier 1 capabilities that contradict each other (shipping both requires mutually exclusive implementation choices)
• Missing "What We Are NOT Doing" section

Pass 2 — INFORMATIONAL

• "Why now" rationale absent or vague
• Key bets missing for a capability
• Open questions section absent
• Tier assignments that seem misaligned with strategic goal

Step 3: Fix-First

Auto-fix without confirmation:

• Missing "What We Are NOT Doing" section: append with [TODO: list explicit deferrals]
• Missing "Open Questions" section: append with questions generated from Pass 1/2 findings

Flag for user decision:

• Capability phrased as deliverable (user must decide the outcome)
• Tier assignment disputes (user decides prioritization)

Step 4: Present review inline

Output the full review report directly in the chat as structured markdown — do not write a file:

## pm:review — {filename} — {date}

**Overall verdict:** Ready | Needs fixes | Blocked

**Findings:** {N} critical, {N} informational

### Critical Issues
{list — each labeled AUTO-FIXED or [needs your decision]}

### Informational
{list — each labeled AUTO-FIXED or noted for awareness}

### Looks good
{specific positives anchored to actual content}

### Auto-fixes applied
{list of edits made directly to the roadmap file}

### Suggested next steps
{ordered list}

Step 5: Interactive resolution (if not --silent)

If not --silent, surface all user-decision items one at a time via AskUserQuestion.

After resolving all items, ask:

AskUserQuestion:
  question: "Roadmap review complete. What next?"
  header: "Next step"
  options:
    - label: "Approve — run pm:approve"
    - label: "Update — run pm:update"
    - label: "Done for now"

Step 6: Notify

if [ "$SILENT" != "true" ]; then
  npx @codevoyant/agent-kit notify \
    --title "pm:review complete" \
    --message "Review complete for {filename}: {N} critical, {N} informational"
fi