AI-Agent-Toolkit spec-agent

Name: spec-agent
Author: ngapngap

Spec Agent Skill

install

source · Clone the upstream repo

git clone https://github.com/ngapngap/AI-Agent-Toolkit

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/ngapngap/AI-Agent-Toolkit "$T" && mkdir -p ~/.claude/skills && cp -r "$T/.agent/skills/spec-agent" ~/.claude/skills/ngapngap-ai-agent-toolkit-spec-agent && rm -rf "$T"

manifest: .agent/skills/spec-agent/SKILL.md

source content

Spec Agent Skill

Overview

Transform debate decisions into actionable specifications and task DAG.

Usage

# Generate spec from debate inputs
node .agent/skills/spec-agent/scripts/generate-spec.js --debate debate.inputs_for_spec.json --intake intake.json

# Generate task breakdown DAG
node .agent/skills/spec-agent/scripts/generate-tasks.js --debate debate.inputs_for_spec.json --output task_breakdown.json

Inputs

```
30_debate/debate.inputs_for_spec.json
```
```
10_intake/intake.json
```
(for constraints reference)

Outputs

```
40_spec/spec.md
```
- Full specification document
```
40_spec/task_breakdown.json
```
- DAG of tasks

spec.md Structure

# spec.md

## Context
[Project context from intake]

## Scope
- MVP: [items from intake.mvp_scope]
- Non-goals: [items from intake.non_goals]

## Decisions (from debate)
- Recommended approach: [chosen_option]
- Reference repos: [if reuse path]
- Key decisions: [from debate.decision]

## Plan & Milestones
[Milestone breakdown]

## Tasks by lane
[Grouped by owner_lane]

## Verification
See qa.md

## Open questions
- Blocking: [must resolve before proceeding]
- Non-blocking: [can resolve during implementation]

Task DAG Rules

Every task has unique
```
node_id
```
```
depends_on
```
references valid node_ids
No cycles in DAG
Each task has
```
owner_lane
```
Each task has
```
validation_cmd
```

Validation

# Validate task_breakdown.json against schema
node .agent/skills/spec-agent/scripts/validate.js --file task_breakdown.json