Domain-specific AI skills for the Valkey ecosystem, authored for AI coding agents and developers.
First release ships one skill: valkey - an application-developer reference. Installable into Claude Code, Codex, Cursor, Cline, GitHub Copilot, OpenCode, Kiro, and 50+ other AgentSkills-compatible agents.
A skill is a reference the agent loads on demand. SKILL.md is a router: trigger phrases and a table of focused topic files. The agent scans the router, reads only the topic file that matches the current task, and leaves the rest out of context. This keeps model attention on your code instead of on generic domain background.
The file format is the AgentSkills open standard. One skill works across 50+ agents unmodified.
32 topic files under skills/valkey/references/, routed from skills/valkey/SKILL.md. Files are split by user-task boundary, so a routed lookup lands in a file that is already the right semantic context:
| Area | Files | Contents |
|---|---|---|
| Features and versions | valkey-features.md, conditional-writes.md, hash-field-ttl.md, commandlog.md, geosearch-bypolygon.md, redis-compatibility.md |
Version-gated commands (IFEQ, DELIFEQ, HSETEX family, COMMANDLOG, GEOSEARCH BYPOLYGON), per-release changelog, Redis migration |
| Application workflows | app-caching.md, app-client-side-caching.md, app-sessions.md, app-locks.md, app-rate-limiting.md, app-queues.md, app-counters-dedup.md, app-leaderboards.md, app-pubsub.md, app-search.md (router: app-patterns.md) |
Caching, CLIENT TRACKING, sessions, locks (Redlock, fencing), rate limiting, queues/streams, counters, leaderboards, pub/sub, search |
| Cluster and HA | cluster-key-client-behavior.md, cluster-slot-migration.md, replication-sentinel-retries.md, durability-persistence.md (router: cluster-and-ha.md) |
Hash tags, CROSSSLOT, MOVED/ASK, slot migration, replication internals, Sentinel, WAIT/WAITAOF, persistence |
| Performance | performance-memory-encoding.md, performance-eviction-ttl.md, performance-fragmentation.md, performance-latency.md, performance-throughput.md, performance-benchmarking.md (router: performance.md) |
Encoding, eviction, fragmentation and defrag, LATENCY + COMMANDLOG diagnosis, pipelining, I/O threads, valkey-benchmark / memtier_benchmark pitfalls |
| Scripting, security, corrections | scripting.md, security.md, anti-patterns.md |
Lua, FUNCTION LOAD, BUSY/KILL rules; AUTH, ACL, TLS; non-obvious corrections and detection commands |
Benchmarking has its own reference (performance-benchmarking.md) covering valkey-benchmark and memtier_benchmark syntax and the pipeline/client-thread pitfalls that skew results.
The authoring lens: what a frontier LLM would not know or would get wrong. Kept are exact defaults, error strings, INFO field names, version-gated behavior, and footguns. For example: DEL is already async-by-default in Valkey 8.0+; IFEQ never creates a missing key; HSET strips field TTL while HSETEX ... KEEPTTL preserves it; PFCOUNT is RW at the key-spec level; BITFIELD caps at u63; valkey-cli --cluster reshard still uses the legacy per-key path in 9.0. Dropped are textbook explanations, per-language client code, and narrative prose.
The valkey skill targets application developers. Server-codebase topics are out of its scope and are planned as separate skills:
valkey-dev- contributing to the Valkey server itself: data structure design and optimization (where memory fetches typically matter more than instruction count), the tcl integration-test framework, unit testing style, and coverage guidelines for changes.valkey-ops- operating Valkey in production: deployment, upgrades, monitoring, capacity.
npx skills add valkey-io/valkey-skills
The skills CLI writes the skill into each targeted agent's expected directory (e.g. .claude/skills/, .cursor/skills/, .codex/skills/, .kiro/skills/). Add -g for global scope or -a <agent> to target specific agents.
/plugin marketplace add valkey-io/valkey-skills
/plugin install valkey
codex plugin marketplace add valkey-io/valkey-skills
codex plugin add valkey@valkey-skills
Pin the marketplace source to a specific commit, branch, or tag with --ref, or pass a local checkout path while testing changes.
Clone the repo and copy skills/valkey/ into the agent's skills directory.
Authoring rules and ground rules for the repo live in CLAUDE.md (which AGENTS.md symlinks to). Key points:
- Plain text, single-dash em-dashes.
- Verify skill content against the actual Valkey source at the declared baseline version; do not rely on web search.
- Keep SKILL.md a grep-optimized trigger list. Reference files are topic-per-file; do not fragment on a 300-line rule.
- No transient files in the repo (summary, plan, audit, analysis).
Issues and PRs welcome.
BSD-3-Clause. See COPYING.