engineering-skills

Turn your AI coding agent into a disciplined engineering team. A reusable, agent-agnostic library of 64 production-grade engineering skills — each a step-by-step process with verification, not reference docs — plus a 38-role SDLC & marketing org and an orchestrated build loop that conducts a feature end to end.

Works with Claude Code, Cursor, Gemini CLI, GitHub Copilot, and Codex / OpenCode. Install once, use in any repo.

How it compares

Most skill libraries are either narrow (the dev loop only) or unproven advice. This one is different on two axes:

• Breadth — the full lifecycle plus a 40-persona org (PM, architect, DBA, QA, SRE, release… and a

marketing team), orchestrated by orchestrated-delivery.

• Rigor — skills are behaviorally tested: every process skill has a pressure-scenario test

that proves the skill changes what the agent does, enforced in CI.

It's new — competing on scope and rigor, not adoption. Honest feedback and PRs welcome.

See it work: a feature shipped end-to-end by the 38-role org (/deliver Add product reviews).

Install (Claude Code plugin — use in any repo)

From inside any project, in an interactive claude session:

/plugin marketplace add vindreshsingh/engineering-skills
/plugin install engineering-skills@engineering-skills

Or wire it into a project declaratively in .claude/settings.json:

{
  "extraKnownMarketplaces": {
    "engineering-skills": {
      "source": { "source": "github", "repo": "vindreshsingh/engineering-skills" }
    }
  },
  "enabledPlugins": {
    "engineering-skills@engineering-skills": true
  }
}

This makes all 64 skills, 38 agents (30 SDLC + 8 marketing), and session hooks available in that project. See docs/plugin-discovery.md for load paths.

Start here (first 5 minutes)

64 skills is a lot — you don't pick from all of them. Get going fast:

1. Run `/start` in any repo. It detects your stack and recommends the 3–5 skills and the workflow

that fit this project — not the whole catalog.

1. Set a profile to scope the emphasis to your kind of work — frontend, backend, full-stack,

or solo-founder (profiles/):

   echo full-stack > .engineering-skills-profile

1. Browse the [live explorer](https://vindreshsingh.github.io/engineering-skills/) when you want to

see everything and copy a ready-to-paste prompt.

See the lift in code: before / after examples — the same task done without a skill vs with it.

Use with any other agent

Works with Claude Code, Gemini CLI, Cursor, GitHub Copilot, and any AGENTS.md-aware tool (Codex, OpenCode, Factory…) — the repo ships the entry-point file each one reads, all pointing at the same skills.

Skills are plain Markdown. Clone the repo and point your agent at skills/<name>/SKILL.md:

git clone https://github.com/vindreshsingh/engineering-skills.git

Per-tool install and usage: docs/platforms.md. General guide: docs/getting-started.md.

How it works

Every task enters through skill-router (or /start), which routes it to the one skill that governs that step — then chains to the next phase. agent-guardrails is always on; skill-harvest feeds lessons back so the library compounds.

flowchart TD
    G["agent-guardrails — always on"]
    S["/start · skill-router"]
    G -.-> S
    S --> DEF

    subgraph LC["Lifecycle — skill-router routes each task to its skill"]
      direction LR
      DEF["Define<br/>product-discovery · spec-first"] --> PLN["Plan<br/>work-planning · threat-modeling"]
      PLN --> BLD["Build<br/>incremental-delivery · test-first"]
      BLD --> VER["Verify<br/>browser-checks · contract-testing · experimentation"]
      VER --> REV["Review<br/>review-gate · hardening · simplify"]
      REV --> SHP["Ship<br/>feature-flags · migration-path · launch-readiness"]
      SHP --> OPS["Operate<br/>observability · incident-response"]
      OPS --> GRW["Grow<br/>launch-campaign · growth-strategy"]
    end

    GRW -.learnings.-> DEF

    subgraph X["Cross-cutting"]
      direction LR
      AUTO["Operate-Autonomously<br/>autonomous-loops · long-running-agents<br/>agent-verification · agent-memory"]
      HARV["skill-harvest → skill-creator"]
    end
    BLD -.-> AUTO
    GRW -.-> HARV

The bookends — product-discovery (/discover) and launch-campaign (/launch) — wrap the code loop into a full discover → build → launch arc.

Skills by phase

See SKILLS.md for the full auto-generated catalog. End-to-end example: docs/sdlc-walkthrough.md.

Explore

A self-contained Agent & Skill Explorer — browse every skill and agent grouped by lifecycle phase, search, and copy a ready-to-paste Claude Code prompt (optionally scoped to a repo/task).

Live: vindreshsingh.github.io/engineering-skills (auto-published from main). Or run it locally:

python3 scripts/generate-explorer.py          # regenerate explorer/index.html from frontmatter
python3 -m http.server 8000 --directory explorer   # then open http://localhost:8000

It's a storefront, not a runtime: the copied prompt runs in Claude Code, under its permissions and this repo's agent-guardrails. The page is generated from the same SKILL.md / agent frontmatter as the catalog, so it never drifts.

Agent personas

Three general reviewer personas:

• code-reviewer — five-dimension review (correctness, readability, architecture, security, performance)
• security-auditor — vulnerability detection, threat modeling, secure-coding review
• test-engineer — test strategy, test writing, coverage analysis

Plus a full 30-role SDLC org and an 8-role growth team for marketing — positioning, content, social, SEO, and community. orchestrated-delivery conducts them across the lifecycle, dispatching independent work to parallel subagents. See docs/agent-org.md for the full org map.

flowchart TD
    OD["orchestrated-delivery<br/>(conductor across phases)"]
    OD --> PRD & ARC & ENG & QA & OPSR & SEC & DOC & GRW

    PRD["Product<br/>product-manager · product-analyst<br/>business-analyst · scrum-master · product-grooming"]
    ARC["Architecture<br/>solution-architect · technical-architect"]
    ENG["Engineering<br/>frontend · backend · database · platform<br/>senior-developer · team-lead · engineering-manager"]
    QA["Quality<br/>qa-engineer · sdet · test-engineer<br/>code-reviewer · technical-qc"]
    OPSR["Ops & Reliability<br/>devops · site-reliability<br/>release-manager · incident-commander"]
    SEC["Security & Risk<br/>security-architect · security-auditor<br/>compliance · risk-assessment · dependency-analyzer"]
    DOC["Design & Docs<br/>ux-designer · technical-writer"]
    GRW["Growth<br/>growth-lead · content-marketer · social-media-manager<br/>seo-strategist · community-manager · email · paid · referral"]

Each persona loads its primary skill and runs that skill's process — the agent is who, the skill is how.

Repository layout

skills/           → Engineering skills (marketing under skills/marketing/)
agents/sdlc/      → SDLC personas (30 agents)
agents/marketing/ → Growth team (8 agents)
marketing/        → Marketing team guide, router, references
hooks/            → Session lifecycle hooks (skill router auto-loaded on session start)
references/       → Engineering checklists (testing, performance, security, accessibility, + phase checklists)
docs/             → Per-tool setup guides
.claude-plugin/   → Plugin + marketplace manifests

Marketing team

A separate Growth & Marketing team for post-launch growth — grouped under marketing/:

• marketing/README.md — team guide and flow
• agents/marketing/ — 5 agents (growth-lead, content-marketer, etc.)
• skills/marketing/ — 5 skills (growth-strategy, content-marketing, etc.)
• prompts/agents/marketing/ — copy-paste prompts

SDLC team

• agents/sdlc/README.md — 30 agents including ux-designer, technical-writer

Authoring new skills

Follow docs/skill-anatomy.md. Every skill needs YAML frontmatter with name + description, and the sections: Overview, When to Use, Process, Common Rationalizations, Red Flags, Verification.

License

MIT — see LICENSE.