46 lines
3.3 KiB
Markdown
46 lines
3.3 KiB
Markdown
# Agent Bootstrap
|
|
|
|
When working in this repository, read this file first.
|
|
|
|
## Start Here
|
|
1. Read `.agents/skills/orchestrator/SKILL.md` before planning or editing.
|
|
2. Use the orchestrator routing rules to choose any specialist skill needed for the task.
|
|
3. Read the selected specialist `SKILL.md` before making decisions, writing code, or reviewing output.
|
|
|
|
## Default Skill Routing
|
|
- Project planning, coordination, delegation, or handoff: `.agents/skills/orchestrator/SKILL.md`
|
|
- React SPA implementation: `.agents/skills/frontend-lead/SKILL.md`
|
|
- React SPA review only: `.agents/skills/frontend-code-auditor/SKILL.md`
|
|
- Next.js SEO, metadata, rendering, or production build standards: `.agents/skills/nextjs-seo-architect/SKILL.md`
|
|
- Backend, API, server, or MCP tools: `.agents/skills/backend-mcp-lead/SKILL.md`
|
|
- Database, Prisma, migrations, schema, indexes, or constraints: `.agents/skills/database-architect/SKILL.md`
|
|
- CMS architecture, page builder, or content model: `.agents/skills/cms-architect/SKILL.md`
|
|
- CMS content management or MCP endpoint testing without code edits: `.agents/skills/cms-content-manager/SKILL.md`
|
|
- Product scope, roadmap, priority, or product gaps: `.agents/skills/product-manager/SKILL.md`
|
|
- UX/UI design, interaction, layout, or visual direction: `.agents/skills/ux-ui-designer/SKILL.md`
|
|
- QA, user flow, regression, empty/error states, or release readiness: `.agents/skills/qa-product-auditor/SKILL.md`
|
|
- Brand, SEO, positioning, or marketing copy: `.agents/skills/brand-strategist/SKILL.md`
|
|
- Design system, color contrast, readability, or token compliance: `.agents/skills/design-system-auditor/SKILL.md`
|
|
- Dynamic Theme compliance, cleaning up hardcoded UI colors: `.agents/skills/theme-compliance-auditor/SKILL.md`
|
|
|
|
## Operating Rules
|
|
- Do not guess field names, API contracts, routes, schema, design tokens, or component names.
|
|
- Read source files, docs, schema, or task context before planning or editing.
|
|
- Do not use silent workarounds to hide contract, type, tool, or schema problems.
|
|
- Escalate blockers when the correct fix belongs to another owner, team, tool, or system layer.
|
|
- For risky frontend work, use `frontend-code-auditor` as a review gate before handoff.
|
|
|
|
## Handoff Expectations
|
|
Every final handoff should summarize:
|
|
- what changed
|
|
- what was verified
|
|
- what files or areas were affected
|
|
- remaining risks or blockers
|
|
- next owner or decision needed
|
|
|
|
## MCP Blackbox Testing Protocol
|
|
- When instructed to test, use, or audit MCP tools, you MUST operate in **Strict Blackbox Mode**.
|
|
- **FORBIDDEN:** You are strictly prohibited from using `view_file`, `grep_search`, `run_command`, or reading any backend source code (e.g., `schema.prisma`, `types`) to find answers or discover capabilities.
|
|
- **ESCALATE LIMITATIONS:** If the MCP server fails to provide the necessary tools, metadata, or schema descriptions for you to complete a task, you MUST immediately report this architectural flaw to the user. Do not use local file system tools as a silent workaround.
|
|
- **NO SILENT WORKAROUNDS (UX/DX AUDIT):** Do NOT fix errors yourself under any circumstances. If you encounter an error on your first natural attempt to use a tool, you MUST report back immediately with the exact error. Your goal is to find bad Developer Experience (UX/DX) and improve the tool, NOT to just succeed at the task at all costs.
|