Production-Ready AI Skill Engineering
awesome-skills
Not just prompt templates. A traceable five-layer system for designing, evaluating, and integrating Claude Code skills into real engineering workflows.
21 Production
Skills
Skills
42 Rationale
Docs
Docs
169 Golden
Fixtures
Fixtures
347x Dev-Time
ROI
ROI
A Five-Layer Knowledge Chain
Each layer feeds the next — from abstract principles to measured, executable evidence.
bestpractice/ Methodology Design principles, evaluation framework, iteration patterns, and workflow integration rationale/ Design Logic Why each skill's gates, structure, and tradeoffs are designed the way they are skills/ Executable Skills 21 production-ready skill artifacts shaped by the methodology above evaluate/ Quantitative Proof Formal review reports with trigger accuracy, ROI, and regression data outputexample/ Real Outputs Actual artifacts from real tasks: PDFs, CI configs, test code, screenshots Quantitative, Not Anecdotal
Trigger accuracy, task performance, and token cost-effectiveness — measured across real scenarios with deterministic regression tests.
go-code-reviewer
347x
developer-time ROI
+36 pp signal-to-noise in subtle defect scenarios
Read report → google-search
+74pp
assertion pass rate
Structured evidence chains vs. unstructured ad-hoc search
Read report → unit-test
+38pp
assertion pass rate
Table-driven tests with boundary and concurrency hunting
Read report → 21 Production-Ready Skills
Each skill ships with paired rationale docs, evaluation reports, and regression tests.
Pipeline Backend Development
go-makefile-writer Design or refactor root Makefiles for Go with consistent fmt / test / lint / build gates git-commit Secret-scan + quality gates before commit, then standardized message generation create-pr High-quality PRs with multiple preflight gates and structured body content go-ci-workflow GitHub Actions CI for Go: Make-driven, locally consistent, caching and layered gates go-code-reviewer Defect-first automated code review — real bugs and regressions, not style nits security-review Exploitability-first review across auth, input, dependencies, and container risks
Testing Testing & Validation
unit-test Table-driven tests with boundary hunting, mapping loss, and concurrency edge cases tdd-workflow Practical TDD with Red→Green→Refactor evidence and risk-path coverage api-integration-test Real runtime config, explicit gates, timeout / retry safety, and failure diagnosis e2e-test E2E tests for key user journeys, balanced for CI integration and reliability fuzzing-test Go fuzz tests with applicability gate — refuses low-value targets upfront thirdparty-api-integration-test External contract validation with explicit run gates and timeout controls systematic-debugging Root-cause analysis required before fixes — no guess-driven debugging loops
Research Search & Research
Docs Documentation & Planning
writing-plans Evidence-backed implementation plans with dependency graphs and post-write review loops update-doc Docs-drift detection and evidence-backed README synchronization readme-generator Project-shape detection and evidence-based generation across service / lib / CLI / monorepo tech-doc-writer Runbooks, API docs, RFCs — type classification, audience analysis, anti-staleness rules
A Complete Engineering Pipeline
Backend skills compose into an end-to-end flow from first line of code to merged PR.
◆
Coding
⬣
Write / Fix Tests
unit-test · tdd-workflow · api-integration-test · e2e-test
⬣
make fmt / lint
go-makefile-writer
⬣
git commit
git-commit
⬣
create PR
create-pr
⬣
CI + Review
go-ci-workflow · go-code-reviewer · security-review
✓
Human Review + Merge
Three Steps to Your First Skill
01
Browse and choose a skill
Find a skill in the grid above, or start with google-search as a straightforward first example. Read the paired rationale docs to understand the design decisions.
02
Copy the skill directory
# Project-level (one project)
cp -r skills/<name> .claude/skills/<name>
# User-level (all projects)
cp -r skills/<name> ~/.claude/skills/<name>03
Claude Code activates it automatically
When the task matches, Claude Code invokes the skill — no manual configuration. For a deeper study, read the full methodology: Best Practice →
Where to Start Reading
- 1 bestpractice/README.md Build the overall picture of the methodology
- 2 rationale/google-search/design.md Read the design rationale for a concrete skill
- 3 skills/google-search/SKILL.md Open the executable skill artifact itself
- 4 evaluate/google-search-skill-eval-report.md Read the quantitative evaluation report