Skip to content

epic: Agent Skills — built-in operational skills, authoring guide, registry distribution #718

@joshrotenberg

Description

@joshrotenberg

Overview

Adopt the Agent Skills specification for extensible, portable operational workflows that compose redisctl's MCP tools. Ship built-in skills for common Enterprise and Cloud operations, provide an authoring guide for custom skills, and distribute via skillet registries.

This replaces the earlier approach of building a bespoke runtime-loaded-prompts system (#644) and hardcoded MCP prompts (#712). Skills are the better path because they're portable across agents (Claude Code, Cursor, Gemini CLI, etc.), require no Rust code changes, and have a standard distribution story via skillet.

Goals

  • Ship built-in operational skills that compose redisctl MCP tools
  • Provide a skill authoring guide so users can create custom workflows
  • Publish skills to a skillet registry for easy discovery and installation
  • Cover the most common Enterprise and Cloud operational scenarios

Sub-Issues

Built-in Enterprise Skills

  • pre-upgrade-health-check — verify cluster readiness before upgrade (get_cluster, list_nodes, get_license, list_alerts, get_database_stats)
  • cluster-capacity-planning — assess capacity and recommend scaling (get_all_nodes_stats, get_all_databases_stats, get_all_shards_stats, get_cluster_stats)
  • cluster-troubleshooting — diagnose reported issues via triage workflow (get_cluster, list_alerts, list_nodes, list_logs, get_database_stats, get_database_endpoints)
  • database-performance-analysis — analyze database performance bottlenecks (get_database_stats, get_shard_stats, list_shards, list_database_alerts)

Built-in Cloud Skills

  • cloud-database-provisioning — guide through creating a subscription and database with best practices
  • cloud-migration-planning — plan migration between environments using Cloud API tools

Infrastructure

  • Determine where skills live in the repo (skills/ at root? crates/redisctl-mcp/skills/?)
  • Skill authoring guide — document available tools, patterns for multi-profile operations, examples
  • Publish skills to a skillet registry
  • Add installation instructions to docs (manual copy, skillet install, etc.)

Skill Format

Each skill follows the Agent Skills specification:

skills/
  pre-upgrade-health-check/
    SKILL.md
  cluster-capacity-planning/
    SKILL.md
    references/
      thresholds.md       # Memory/CPU warning thresholds, scaling heuristics
  cluster-troubleshooting/
    SKILL.md
  database-performance-analysis/
    SKILL.md

Example SKILL.md:

---
name: pre-upgrade-health-check
description: >
  Verify a Redis Enterprise cluster is ready for upgrade by checking
  node health, license status, alerts, and database backup state.
  Use when planning a cluster upgrade or maintenance window.
compatibility: Requires redisctl MCP server with enterprise tools enabled
metadata:
  author: redis-developer
  version: "1.0"
---

# Pre-Upgrade Health Check

## When to use
Use this skill before any planned Redis Enterprise cluster upgrade
or maintenance window...

## Steps
1. Check cluster status and version with `get_cluster`
2. Verify all nodes are healthy with `list_nodes` and `get_node`
3. Check license expiry with `get_license` and usage with `get_license_usage`
4. Review active alerts with `list_alerts`
5. Verify database health and backup recency with `list_databases` / `get_database_stats`

## Decision criteria
- **Block upgrade** if: any node unhealthy, license expires within 7 days, critical alerts active, any database backup older than 24 hours
- **Warn** if: license expires within 30 days, non-critical alerts active, memory utilization > 80%
- **Clear** if: all checks pass

## Output
Provide a go/no-go summary table with each check's status and details.

Distribution

  1. Shipped in repo: skills live under skills/ and can be copied to agent skill directories
  2. Skillet registry: skillet install redis/pre-upgrade-health-check
  3. Docs: installation instructions for each supported agent

What stays the same

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

References

Metadata

Metadata

Assignees

No one assigned

    Labels

    epicParent issue tracking multiple related issuesmcpRelated to the MCP server

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions