# Executive Claude Creation Guide

Use this guide with **https://mock.netray.co/sample**.

## Choose the right extension

### Create a Skill when

- the same decision checklist or investigation method is repeatedly pasted into Claude;
- the executive expects a consistent evidence standard or output format;
- the workflow should load only when relevant;
- detailed references can stay outside the main `SKILL.md` until needed.

Recommended project location:

```text
.claude/skills/<skill-name>/SKILL.md
```

Add references, examples, and deterministic scripts beside `SKILL.md` when useful. Write a precise description, define trigger and non-trigger conditions, accept `$ARGUMENTS`, discover MCP tools before using them, cite source IDs, label uncertainty, and stop for approval before consequential writes.

Official guide: https://code.claude.com/docs/en/slash-commands

### Create a Plugin when

- the capability should be versioned and shared with a team;
- several Skills work together;
- MCP configuration should travel with the package;
- a focused agent or deterministic validation hook is justified;
- administrators need install, verification, upgrade, and rollback guidance.

```text
your-plugin/
├── .claude-plugin/
│   └── plugin.json
├── skills/
├── agents/          # optional
├── hooks/           # optional
├── .mcp.json
├── README.md
└── tests/
```

Only `plugin.json` belongs inside `.claude-plugin/`. Use `${CLAUDE_PLUGIN_ROOT}` for Plugin-relative paths. Do not embed secrets.

Official guides:

- https://code.claude.com/docs/en/plugins
- https://code.claude.com/docs/en/plugins-reference

### Create an MCP server when

- Claude needs governed evidence or an action that is not already exposed;
- atomic tools with clear schemas would be more reliable than open-ended browsing;
- record-level provenance, structured errors, rate limits, auditability, and approval boundaries matter.

Start read-only with synthetic data. Add README/schema/example resources, reusable MCP prompts, strict validation, stable JSON, bounded outputs, tests, and an operational runbook.

Official guide: https://code.claude.com/docs/en/mcp

## Executive build standard

Every shared capability should:

1. Discover capabilities and schemas rather than guessing.
2. Identify the decision and evidence needed.
3. Cite stable mock record IDs.
4. Expose conflicts and competing explanations.
5. Distinguish observed, calculated, inferred, scenario, and recommendation.
6. State confidence, freshness, and missing evidence.
7. Protect personal, payroll, assessment, and customer information.
8. Pause for human approval before consequential writes.
9. Validate output and record privacy-safe audit metadata.
10. Include install, test, upgrade, rollback, and support guidance.

## Hosted workshop MCPs

Excelsoft Analytics, Excelsoft Books, Excelsoft CRM, SARAS Test & Assessment, Excelsoft People, Support & Exam-Day Ops, AI-Levate, SARAS LMS, Excelsoft HRMS, and Excelsoft ERP.

All hosted data is synthetic. Never place AWS, Cloudflare, database, application, or production credentials in prompts, Skills, Plugin files, MCP configuration, examples, or email.

Managed by Netray · https://netray.co