Skip to content
Back to skills

Feature Spec Review Panel

ASecurity

Multi-agent Feature Spec review from 7 perspectives (engineering, design, executive, legal, UX research, skeptic, customer voice). Spawns parallel sub-agents for comprehensive feedback.

  • 4 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added May 26, 2026
ai-agentstestingdevopssecurityperformance

Security analysis

A100/100

Scanned May 27, 2026

npx -y skills add ramybarsoum/prodkit --skill feature-spec-review-panel --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Feature Spec Review Panel?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Feature Spec Review Panel
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/ramybarsoum-feature-spec-review-panel/badge)](https://www.skillsdirectory.com/skills/ramybarsoum-feature-spec-review-panel)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
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.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.