feat(mcp): ship built-in Agent Skills and skill authoring guide

[joshrotenberg]

Summary

Add support for user-defined workflow prompts that can be saved as markdown files and invoked through MCP.

Updated direction: Rather than building a bespoke runtime-loaded-prompts system, adopt the Agent Skills specification for extensible workflows. Ship built-in skills with redisctl and provide a guide for users to author their own.

Why Agent Skills instead of custom workflows

The original design proposed a ~/.config/redisctl/workflows/ directory with markdown files loaded by the MCP server at runtime using MCP's prompts feature. Since then, the Agent Skills spec has emerged as a cross-agent standard supported by Claude Code, Cursor, Gemini CLI, Copilot, and others.

Skills are the better path because:

• Portable: Skills work across any skills-compatible agent, not just redisctl's MCP server
• No code changes required: Skills are SKILL.md files that compose existing MCP tools
• Standard format: YAML frontmatter + markdown instructions, already supported by major agents
• Progressive disclosure: Agents load only name/description at startup, full instructions on activation
• Distribution via skillet: Skills can be published to and installed from skill registries

Scope

1. Ship built-in enterprise skills

Create SKILL.md files for the operational workflows described in #712:

• pre-upgrade-health-check -- verify cluster readiness before upgrade
• cluster-capacity-planning -- assess cluster capacity and recommend scaling
• cluster-troubleshooting -- diagnose reported issues via triage workflow
• database-performance-analysis -- analyze database performance bottlenecks

Each skill references redisctl MCP tools by name and provides step-by-step instructions for the agent to follow.

2. Skill authoring guide

Document how users can create custom skills that compose redisctl's MCP tools:

• Available tool reference (names, inputs, outputs)
• Patterns for multi-profile operations (e.g., "for each Enterprise profile...")
• Examples of parameterized workflows
• How to install skills into agent directories or publish to a skillet registry

3. Distribution

Skills can be distributed via:

• Shipped in the redisctl repo (installed alongside the tool)
• Published to a skillet registry for skillet install redis/pre-upgrade-health-check

What stays the same

The existing hardcoded MCP prompts in prompts.rs (troubleshoot, performance, capacity planning, migration) remain as-is. They serve a different purpose via the MCP prompts/list protocol.

References

• Agent Skills specification: https://agentskills.io/specification
• Example skills: https://github.com/anthropics/skills
• Skillet (skill registry toolkit): https://github.com/joshrotenberg/grimoire
• Enterprise operational skills: feat(mcp): create enterprise operational Agent Skills #712
• Enterprise epic: epic: Enterprise — operational workflows, cluster health, coverage #679

[joshrotenberg]

Superseded by #718. Rather than a bespoke runtime-loaded-prompts system, adopting the Agent Skills specification for portable, cross-agent workflows. Skills compose existing MCP tools and are distributed via skillet registries.