claude/ai-resources-plugin/skills/add-design-agent/SKILL.md
Create a research or review agent for the /design workflow.
npx skillsauth add amhuppert/my-ai-resources add-design-agentInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Create a new project-specific design agent that integrates with the /design workflow.
-agent)Extract both values from the user's invocation input.
If the agent type is missing or is not "research" or "review", ask the user with the active client's user-question mechanism:
Use the active client's user-question mechanism to gather:
Question 1: Agent Name
Question 2: Domain/Expertise
Question 3: Key Responsibilities
Create the agent directly using the active client's custom-agent format. If a dedicated agent-development skill is available, use it.
Provide this context to the skill:
Create a design workflow agent with these specifications:
**Type**: [research|review]
**Name**: [agent-name]-agent
**Domain**: [user's expertise description]
**Responsibilities**: [user's selected/described responsibilities]
**Requirements**:
1. Frontmatter must include:
- name: [agent-name]-agent
- description: Use this agent when [appropriate triggering conditions based on type and domain]
- model: sonnet
- color: [choose appropriate color - not already used by universal agents: magenta, gray, blue, red, cyan, yellow, orange]
- tools: Read, Grep, Glob, Write [add WebSearch, WebFetch if research agent]
2. System prompt must include:
- Clear role description as [research expert | reviewer] in [domain]
- Core responsibilities (3-5 items based on user input)
- Process/methodology for [researching | reviewing]
- Output format following design workflow conventions:
For RESEARCH agents:
```markdown
# [Agent Name] Research
## Executive Summary
[2-3 sentences on key findings]
## Key Findings
### [Finding 1]
- Observation
- Implications
- Recommendation
## Recommendations
[Prioritized list]
## Tradeoffs and Alternatives
[Options considered, why recommendations chosen]
## References
[Sources consulted]
```
For REVIEW agents:
```markdown
# [Agent Name] Review
## Summary
[1-2 sentence assessment]
## Critical Issues
### [Issue 1]
- Location: [where]
- Problem: [what]
- Impact: [why it matters]
- Recommendation: [how to fix]
## Major Issues
[Same format]
## Minor Issues
[Same format]
## Suggestions
[Nice-to-haves]
## What's Working Well
[Positive aspects]
```
3. Save the agent in the active client's project agent directory:
- Claude Code: `.claude/agents/[agent-name]-agent.md`
- Codex: `.codex/agents/[agent-name]-agent.toml`
Save the generated agent file to the active client's path above. Create the parent directory if it doesn't exist.
Check if memory-bank/DESIGN-AGENTS.md exists.
If it doesn't exist:
references/design-agents-template.mdIf it exists:
- [agent-name]-agent: [brief description of expertise]Report to the user:
## Design Agent Created
**Agent**: [agent-name]-agent
**Type**: [Research|Review]
**Location**: [the active client's project agent path]
**Added to**: memory-bank/DESIGN-AGENTS.md
### Next Steps
1. Review the agent at its reported location
2. Customize the system prompt if needed
3. Run `/design` to use the new agent in your design workflow
The agent will automatically be included in:
- [If research]: Phase 2 (Research) of new design workflows
- [If review]: Phase 4 (Review) of both new design and iteration workflows
development
Debug a running web app via the web-debugger SDK: app logs, application state, runtime snapshots, React state, query cache.
development
Thoroughly understand a software development objective before implementation: research, identify ambiguities, ask clarifying questions. Use before starting implementation of a non-trivial or ambiguously specified feature, or when requirements leave open design decisions.
development
Locate the on-disk Claude Code transcript file (.jsonl under ~/.claude/projects/) for the current or a specified conversation.
development
Reflect on codebase navigation effectiveness at end of conversation. Surfaces dead ends, inefficiencies, missing context. Does not write files — pair with /kiro:steering-custom to persist.