skills

Skills for hardcore engineering:

• pragmatic-architecture/ — practical design principles: small interfaces, locality, clear dependencies, validation, and behavioral testing, balanced against effort, migration risk, and existing project conventions.
• pragmatic-testing/ — behavior-focused testing through realistic boundaries, reuse of test services and API clients, integration tests, and regression evidence.
• codeps/ — measure the dependency graph before refactoring with Codeps: generate package/file health status from SemanticDB, jdeps, or a generic export; diagnose cycles, change propagators, public surface, mutable state, structural use, and orphan candidates; compare health and findings through commits, then verify refactorings on a fresh build and report.
• process-hygiene/ — startup/shutdown, signals, cancellation, and ownership of tasks, threads, executors, and subprocesses. Includes JVM and shell references.
• cli-design/ — terminal interfaces, flags, help, output contracts, exit codes, interactivity, and CLI configuration conventions.
• twelve-factor-app/ — deployment configuration, backing services, stateless instances, releases, parity, platform logs, and admin tasks.

Install

Claude Code, Cursor, Codex, OpenCode...

npx skills add sake92/skills -g

Omit -g to install project-local. Use --agent <name> to target a specific agent if it's not auto-detected.

Evaluations

Evaluation tooling lives under tests/. See individual skill READMEs for evaluation status, example results, and related skills.

Unless stated otherwise, evaluation examples use Pi 1.0.4 with GPT-6 Luna at low reasoning, with one candidate per configuration. Reported time covers candidate execution, excluding grading; tokens count input and output, excluding cached input.

Skill	Without skill	With skill	Score gain
Pragmatic architecture	26/35 (74.3%)	35/35 (100%)	+25.7 pp
Process hygiene	23/29 (79.3%)	28/29 (96.6%)	+17.2 pp
Twelve-Factor app	7/8 (87.5%)	8/8 (100%)	+12.5 pp
Codeps	9/13 (69.2%)	12/13 (92.3%)	+23.1 pp
Pragmatic testing	12/17 (70.6%)	16/17 (94.1%)	+23.5 pp
CLI design	5/9 (55.6%)	9/9 (100%)	+44.4 pp

Scores pool passed checks across the published examples; pp means percentage points. These are selected single-run examples, some developed adaptively, not overall effectiveness estimates.