jars-demo/jars-skills

jars-skills
The agent skills I use most in my day-to-day work, shared in case they help you too.
I'm Jishanahmed AR Shaikh (Mr. JARS). I build and contribute to open source with coding agents
most days, and when I notice I'm explaining the same workflow to an agent again and again, I
write it down here as a skill.
Each skill is one folder with a SKILL.md: a short playbook that teaches an agent to do one job
the way I do it. They follow the Agent Skills format, so they work in
Claude Code, Cursor, GitHub Copilot, Codex, Gemini CLI and other agents that read it.
I add them one at a time, aiming for one a day, and only after I've used the workflow for real.
Install
**Claude Code** (plugin marketplace):
Or in one step (Claude Code 2.1.275+): /plugin install jars-skills --marketplace jars-demo/jars-skills.
Skills then load on their own when a task matches, or run one directly, e.g.
/jars-skills:commit-style. Update later with /plugin marketplace update jars-skills.
**Any agent** (via the skills CLI):
**By hand:** copy a folder from skills/ into your agent's skills directory, for example
~/.claude/skills/commit-style/.
Skills
| # | Skill | Use it when |
| --- | --- | --- |
| 39 | env-config | Designing a typed, validated configuration system: centralises all env var reads into one config module, validates and fails fast at startup with a clear error, distinguishes secrets from plain config, maintains a .env.example with every key documented, and prevents os.environ / process.env scattered across the codebase. |
| 38 | code-structure | Structuring and naming a codebase for long-term navigability: enforces a 696-line file ceiling, splits files by single responsibility, applies snake_case / PascalCase / SCREAMING_SNAKE consistently, wires structured logging with full error context at every boundary, adds file-level navigation headers, and scaffolds a clean folder hierarchy where every directory has exactly one job. |
| 37 | concurrency-and-locking | Identifying and fixing race conditions, deadlocks, and concurrency bugs: establishes happens-before relationships, chooses the right primitive (mutex, rwlock, atomic, channel), detects and breaks deadlock cycles with lock ordering, fixes TOCTOU races, uses CAS loops and copy-on-write atomics, pads to cache lines to eliminate false sharing, and runs the race detector. |
| 36 | query-optimization | Diagnosing and fixing slow database queries: reads EXPLAIN ANALYZE output node by node, identifies sequential scans, cardinality misestimates, and bad join orders, creates the right index type (B-tree, partial, covering, GIN, BRIN), rewrites N+1s, correlated subqueries, implicit casts, and OFFSET pagination, and forces statistics refresh when the planner is wrong. |
| 35 | capacity-planning | Modelling load, projecting resource ceilings, and deciding when to scale: derives RPS and data-volume limits from first principles, identifies the bottleneck resource (CPU, memory, DB connections, I/O) with back-of-envelope math, calculates time-to-saturation, designs load tests that reflect real traffic shape, applies Little's Law to size thread and connection pools, and translates findings into concrete scaling decisions. |
| 34 | memory-and-allocations | Profiling and fixing memory leaks, excessive heap allocations, and GC pressure: captures heap snapshots and allocation timelines, identifies retaining paths and dominator objects, fixes event-listener accumulation, closure captures, unbounded caches, and goroutine leaks, and applies the same workflow across Node.js, Python, Go, and JVM runtimes. |
| 33 | data-consistency | Reasoning about and implementing data consistency in distributed systems: picks the right consistency model (linearisable, causal, eventual), chooses a replication strategy, applies the right isolation level, resolves conflicts with LWW/vector clocks/CRDTs, prevents split-brain with fencing tokens, and makes deliberate CAP and PACELC trade-offs. |