Practical engineering knowledge for coding agents: review a message-processing contract, investigate Linux latency, choose a storage engine, or test an architecture decision.
Each skill starts with a short decision procedure. Deeper mechanisms, original examples, failure cases, and exercises live in references loaded only when needed. The library follows the Agent Skills format.
Pick the skills you need:
npx skills add karlem/skills --skill distributed-systems linux-performanceOr browse the available skills:
npx skills add karlem/skills --listFor a local install without Node.js:
git clone https://github.com/karlem/skills.git
cd skills
python3 scripts/install.py --dest ~/.agents/skills --skill distributed-systems linux-performanceUse --all instead of --skill … to install the library. The installer copies
complete skill folders into a flat agent root and refuses to overwrite existing
skills. Add --dry-run to preview. If the repository is private, authenticate
GitHub before cloning.
| Agent | Personal skill root |
|---|---|
| Codex | ~/.agents/skills |
| Claude Code | ~/.claude/skills |
| GitHub Copilot CLI | ~/.copilot/skills |
| OpenCode | ~/.config/opencode/skills |
Pass the matching root to --dest, then start a new agent session. These are
portable Markdown skills: no agent plug-in, background service, model key, or
runtime package is required. Format checks cover these hosts; a live discovery
test in every host has not been performed. See usage for
project-local installation and examples.
| Skill | Use it for |
|---|---|
| distributed-systems | Replication, consistency, crash windows, ordering, deduplication |
| linux-performance | p99 investigations, USE, pressure, quotas, benchmark design |
| systems-profiling | CPU versus waiting time, flame graphs, profile interpretation |
| architecture-tradeoffs | Service boundaries, transaction placement, migration decisions |
| database-internals | B-tree/LSM trade-offs, amplification, WAL and checkpoints |
| systems-programming | C arithmetic, memory lifetime, partial I/O, locality |
| production-reliability | Deadlines, retries, bulkheads, backpressure, recovery |
| concurrent-programming | Races, histories, publication, progress, reclamation |
| architecture-patterns | Quality scenarios, architecture selection, fitness checks |
| design-patterns | Variation, behavioral contracts, compatible product families |
| network-engineering | RTT/BDP budgets, transport contracts, forwarding and path diagnosis |
Ask your agent, for example:
Use distributed-systems to review this consumer. Show crash windows, the exact invariant, and a replay-safe implementation before recommending changes.
Use linux-performance to investigate these p99 spikes. Give competing hypotheses, the next measurements, and a controlled validation plan.
All ten requested books have a coverage record. Coverage is partial for every book. No authorized full edition was available; the skills use official samples, author materials, research papers, standards, and implementation documentation. They contain original engineering procedures rather than chapter reproductions.
- Book coverage: editions, available material, precise gaps.
- Source manifest: URLs, access status, dates, hashes, extraction methods, and acquisition issues.
- Conversion workflow: verified use of book-to-skill, its limits, and the supplementary extraction path.
python3 scripts/validate.py
python3 -m unittest discover -s tests -vThe checks cover structure, local links, provenance, installation behavior, and selected engineering mechanisms. Evaluation scenarios include prompts, acceptance criteria, reviewed responses, and executable checks. They do not establish a measured improvement in model performance.
See the completion report and execution log for results and remaining limits. Contributions are welcome through pull requests; read CONTRIBUTING.md.
MIT applies to original repository content. Source books and linked materials retain their own rights and licenses. No books, extracted source text, slides, or publisher diagrams are included.