Engineering Skills

中文 · MIT · CI

Engineering workflows for Claude Code, Codex and other agents that read Agent Skills. Use task-specific guidance for requirements, design, implementation, verification and operation, with concise shared rules and human-auditable documents.

Install

Claude Code

/plugin marketplace add https://github.com/int2t05/engineering-skills
/plugin install engineering-skills@int2t05

Codex — managed plugin:

codex plugin marketplace add https://github.com/int2t05/engineering-skills.git
codex plugin add engineering-skills@int2t05

Update with codex plugin marketplace upgrade int2t05, then codex plugin add engineering-skills@int2t05 to refresh the installed copy. Restart Codex if needed.

Standalone skills / other hosts — Python 3.10 or later, including Windows:

git clone https://github.com/int2t05/engineering-skills.git
cd engineering-skills
python scripts/install-skills.py

Installs self-contained skill folders into ~/.agents/skills. Shared resources are bundled; no plugin root variable or symlink is needed. Use --skill implement --skill debugging for a selection. Installation and updates covers conflicts, local edits and other destinations. Avoid duplicate installations in the same host.

Use

Claude Code: /implement. Codex plugin: $engineering-skills:implement; standalone: $implement. Supported hosts may also select skills from matching intent. brainstorm and handoff require explicit invocation. Use using-skills when routing is uncertain, not before every simple task.

Start at the task that needs solving; a clear edit does not need a full lifecycle. Confirm document paths before creation, movement or renaming. Significant requirements and design remain reviewable; human acceptance is separate from technical checks.

Task prompts · Routing · Output choices · Shared rules

Catalog

Phase	Skills
01-product	brainstorm, oss-strategy, spec
02-research	research
03-design	api-design, architecture, codebase-design, design-research, domain-modeling, frontend-design, image-to-code, imagegen, prompt-engineering, prototype, schema-design
04-develop	auth-implementation, breakdown, context-engineering, error-handling, i18n, implement, multi-agent-orchestration
05-tune	cost-optimization, performance, refactoring, simplify
06-test	api-testing, e2e-testing, load-testing, tdd, test-generation
07-verify	a11y-review, code-review, debugging, linting, security-review
08-ship	ci-cd, deprecation-migration, git-workflow, oss-polish, shipping
09-operate	documentation-audit, handoff, incident-response, observability, purity-audit
meta	skill-authoring, using-skills

Verify and contribute

python scripts/gen-agents-yaml.py
bash scripts/validate-skills.sh
python -m unittest discover -s test -v

CI checks schema, catalogs, output declarations, adapter drift and self-contained installation. Behavioral model evaluation is separate and on demand; these checks do not prove model quality. See evaluation guidance. Report a reproducible problem via GitHub Issues; the issue and PR templates identify scope and evidence. Follow skill anatomy when contributing a workflow.

Layout

• skills/: phase-organized workflows, references and generated Codex adapters.
• references/: shared engineering, product, design and human-audit guidance.
• scripts/ and test/: validation, installation and real filesystem checks.
• .claude-plugin/ and hooks/: Claude Code integration.
• docs/: task prompts, output guidance and project coordination.

Sources

Workflow comparisons: OpenAI skills, Anthropic skills, Superpowers, Addy Osmani and Matt Pocock. Instructions are adapted to this pack's task boundaries and approval model; reference projects are not runtime dependencies.