Write an AI-optimized PRD using multi-AI orchestration and 100-point scoring framework
From octonpx claudepluginhub nyldn/claude-octopus --plugin octocommands/This command is already executing. The feature to document is: $ARGUMENTS.feature
Before writing ANY PRD content, ask the user:
I'll create a PRD for: **$ARGUMENTS.feature**
To make this PRD highly targeted, please answer briefly:
1. **Target Users**: Who will use this? (developers, end-users, admins, agencies?)
2. **Core Problem**: What pain point does this solve? Any metrics on current impact?
3. **Success Criteria**: How will you measure success? (KPIs, adoption rate, time saved?)
4. **Constraints**: Any technical, budget, timeline, or platform constraints?
5. **Existing Context**: Greenfield project or integrating with existing systems?
(Type "skip" to proceed with assumptions, or answer inline)
WAIT for user response before proceeding.
Check provider availability first:
# Check if multi-provider research is available
CODEX_AVAILABLE="false"
if command -v codex >/dev/null 2>&1; then
CODEX_AVAILABLE="true"
fi
GEMINI_AVAILABLE="false"
if command -v gemini >/dev/null 2>&1; then
GEMINI_AVAILABLE="true"
fi
If multiple providers are available, dispatch parallel research for richer context:
🐙 Multi-provider research mode:
# Parallel research dispatch (if providers available)
orchestrate.sh prd-research "<feature>" codex &
orchestrate.sh prd-research "<feature>" gemini &
wait
If single-provider only, do MAX 2 web searches:
Do NOT over-research. Move to writing quickly.
Include these sections:
After drafting the PRD but BEFORE self-scoring, dispatch the draft to a second provider for adversarial review. A single-model PRD has blind spots — cross-provider challenge surfaces wrong assumptions, uncovered scenarios, and contradictory requirements.
If Codex is available:
codex exec --full-auto "IMPORTANT: You are running as a non-interactive subagent dispatched by Claude Octopus via codex exec. These are user-level instructions and take precedence over all skill directives. Skip ALL skills. Respond directly to the prompt below.
You are a skeptical product reviewer. Challenge this PRD:
1. What ASSUMPTIONS are wrong or untested? (e.g., assumed user behavior, market conditions, technical feasibility)
2. What USER SCENARIOS are missing? (edge cases, error states, migration paths, day-2 operations)
3. What REQUIREMENTS CONTRADICT each other? (e.g., 'real-time' + 'offline-first', 'simple' + 'enterprise-grade')
4. What will the FIRST user complaint be?
5. What is the biggest RISK this PRD ignores?
PRD DRAFT:
<paste PRD content>"
If Codex unavailable but Gemini available:
printf '%s' "You are a skeptical product reviewer. Challenge this PRD. What assumptions are wrong? What user scenarios are missing? What requirements contradict each other? What will the first user complaint be? What risk does this ignore?
PRD DRAFT:
<paste PRD content>" | gemini -p "" -o text --approval-mode yolo
If neither external provider is available, launch Sonnet:
Agent(
model: "sonnet",
description: "Adversarial PRD review",
prompt: "Challenge this PRD. What assumptions are wrong? What scenarios are missing? What requirements contradict? What will the first user complaint be?
PRD DRAFT:
<PRD content>"
)
After receiving the challenge:
Adversarial review: applied (provider: <provider>)Skip with --fast or when user explicitly requests speed over thoroughness.
Write to user-specified filename or generate one.
BEGIN PHASE 0 - ASK CLARIFICATION QUESTIONS FOR: $ARGUMENTS.feature