Multi-agent Feature Spec review from 7 perspectives (engineering, design, executive, legal, UX research, skeptic, customer voice). Spawns parallel sub-agents for comprehensive feedback.
Scanned 5/27/2026
Install via CLI
openskills install ramybarsoum/prodkit---
name: feature-spec-review-panel
description: Multi-agent Feature Spec review from 7 perspectives (engineering, design, executive, legal, UX research, skeptic, customer voice). Spawns parallel sub-agents for comprehensive feedback.
allowed-tools: Agent, Read, Glob, Grep, AskUserQuestion, TodoWrite
---
> **Note:** This skill is an alias for `feature-doc-review-panel`. "Feature Spec" and "Feature Doc" are treated identically. The canonical term is "Feature Doc" but this skill exists for backward compatibility.
> **Interaction style:** Use the `AskUserQuestion` tool for all structured questions in this skill. Group related questions together (2-3 per call) rather than asking one at a time.
## Purpose
Get comprehensive feedback on your Feature Spec/Doc from 7 different perspectives in parallel: Engineering, Design, Executive, Legal, UX Research, Skeptic, and Customer Voice.
Catches gaps, challenges assumptions, and surfaces conflicts before stakeholder review.
## Usage
- `/feature-spec-review-panel` - Review a Feature Spec with all 7 sub-agents
- `/feature-spec-review-panel [file-path]` - Review specific Feature Spec by path
- `/feature-spec-review-panel --perspectives "eng,design,exec"` - Review with subset of agents
- `/feature-spec-review-panel --extended` - Include 6 additional perspectives (architecture, data, devops, performance, security, testing)
## Execution Steps
### Step 1: Locate the Feature Spec/Doc
If a file path was provided as an argument, use that. Otherwise:
1. Search for Feature Specs/Docs in the workspace: `Glob("**/FEATURE-DOC.md")`, `Glob("**/feature-doc*.md")`, `Glob("**/feature-spec*.md")`, `Glob("projects/**/FEATURE-DOC.md")`
2. If multiple found, ask the user which one to review using `AskUserQuestion`
3. If none found, ask the user to provide the path or paste the content
Read the full document content. Store it as `DOC_CONTENT`.
### Step 2: Load context files (if they exist)
Attempt to read these context files. Skip any that don't exist:
- `reference/company/product-principles.md` (product principles to evaluate against)
- `reference/company/business-info.md` (company context)
- `reference/strategy/` (any strategy docs)
Combine available context into `CONTEXT_SUMMARY` (keep it under 500 words).
### Step 3: Determine which perspectives to run
**Default 7 (core panel):**
| Perspective | Sub-Agent File | Key |
| ------------ | ------------------------------------------ | ---------- |
| Engineering | `_system/sub-agents/engineer-reviewer.md` | `eng` |
| Design | `_system/sub-agents/designer-reviewer.md` | `design` |
| Executive | `_system/sub-agents/executive-reviewer.md` | `exec` |
| Legal | `_system/sub-agents/legal-advisor.md` | `legal` |
| UX Research | `_system/sub-agents/uxr-analyst.md` | `uxr` |
| Skeptic | `_system/sub-agents/skeptic.md` | `skeptic` |
| Customer Voice | `_system/sub-agents/customer-voice.md` | `customer` |
**Extended 6 (with --extended flag):**
| Perspective | Sub-Agent File | Key |
| ------------- | --------------------------------------------- | ---------- |
| Architecture | `_system/sub-agents/architecture-reviewer.md` | `arch` |
| Data | `_system/sub-agents/data-reviewer.md` | `data` |
| DevOps | `_system/sub-agents/devops-reviewer.md` | `devops` |
| Performance | `_system/sub-agents/performance-reviewer.md` | `perf` |
| Security | `_system/sub-agents/security-reviewer.md` | `security` |
| Testing | `_system/sub-agents/testing-reviewer.md` | `testing` |
If `--perspectives` flag is provided, filter to only the specified keys.
### Step 4: Read sub-agent persona files
Read each required sub-agent file from `_system/sub-agents/`. Store the content for each.
### Step 5: Spawn parallel review agents
**CRITICAL: Spawn ALL selected agents in a SINGLE message using multiple Agent tool calls.** This runs them in parallel for speed.
For each perspective, use the `Agent` tool with:
```text
Agent({
description: "[Perspective] review of Feature Spec",
prompt: `You are reviewing a Feature Spec. Adopt the following persona and review framework:
---
[PASTE THE FULL SUB-AGENT PERSONA FILE CONTENT HERE]
---
## Company Context
[PASTE CONTEXT_SUMMARY HERE]
## Feature Spec to Review
[PASTE DOC_CONTENT HERE]
## Your Task
Review this Feature Spec thoroughly from your perspective. Structure your review as:
1. **Verdict**: PASS / NEEDS WORK / BLOCK (one word)
2. **Strengths** (2-4 bullets): What's done well
3. **Issues** (prioritized list): Each with severity (Critical / Major / Minor), description, and recommended fix
4. **Questions**: Open questions that need answers before implementation
5. **Score**: Rate 1-10 from your perspective
Keep your review focused and actionable. No filler. Under 500 words.`
})
```
### Step 6: Synthesize results
After all agents return, compile the unified review:
```markdown
# Feature Spec Review Panel
**Document:** [document name/path]
**Date:** [today]
**Perspectives:** [list of perspectives used]
## Summary Verdict
| Perspective | Verdict | Score | Critical Issues |
| ----------- | ------- | ----- | --------------- |
| Engineering | [PASS/NEEDS WORK/BLOCK] | [X/10] | [count] |
| Design | ... | ... | ... |
| ... | ... | ... | ... |
**Overall:** [PASS if all pass, NEEDS WORK if any need work, BLOCK if any block]
**Average Score:** [X/10]
## Critical Issues (must fix before implementation)
[List all Critical-severity issues from all perspectives, grouped by theme]
## Major Issues (should fix)
[List all Major-severity issues, grouped by theme]
## Conflicting Feedback
[Flag any cases where two perspectives contradict each other]
## Consensus Strengths
[Items praised by 3+ perspectives]
## Open Questions
[Consolidated list of all open questions, deduplicated]
## Recommended Next Steps
1. [Prioritized action items]
2. ...
```
### Step 7: Present and offer to save
Present the synthesized review to the user. Then ask:
> "Want me to save this review? I can write it to `[project-folder]/reviews/spec-review-[date].md`."
If yes, save the file.
## Notes
- Each sub-agent runs independently. They don't see each other's feedback.
- The synthesis step is where conflicts get surfaced and themes emerge.
- For AllCare projects, the Legal and Security perspectives are especially important due to HIPAA requirements.
- If `reference/company/product-principles.md` exists, every perspective should evaluate alignment with those principles.
No comments yet. Be the first to comment!