frontend-eng-team

agent
Guvenlik Denetimi
Uyari
Health Uyari
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 8 GitHub stars
Code Uyari
  • process.env — Environment variable access in scripts/collect-pr-metrics.ts
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

Multi-agent pipeline for AI-assisted frontend engineering. Six roles, human-gated, tool-neutral. Runs in Claude Code, GitHub Copilot, and any AGENTS.md-aware tool.

README.md

Relay

A six-role multi-agent pipeline for shipping frontend changes with AI assistance — tool-neutral, cache-efficient, and gated by humans at the right moments.

CI
License: MIT

Note: the GitHub repo slug is frontend-eng-team for historical reasons. The project name is Relay.

flowchart LR
    T([Task]) --> PM[PM]
    PM --> A[Architect]
    A --> G{Human<br/>approval}
    G -->|approve| I[Implementer]
    I --> R[Reviewer]
    R --> P[Prod-Readiness]
    P --> V[Verifier]
    R -.-> V
    V --> O([PR ready])

    classDef conditional stroke-dasharray: 4 4
    class PM,P conditional

Dashed boxes are conditional. PM runs only for feature work (skipped for bugfix / refactor / chore / docs). Prod-Readiness fires only when the Architect tags the plan sensitive:* (auth, payments, PII, etc.).

Why

  • One context, many roles. Each phase runs under a focused contract (plan, implement, review, verify) instead of one agent doing everything badly.
  • Cache-friendly. Role prompts are stable prefixes; only the task payload changes. Expect >90% cache hit on a warm session.
  • Fails closed where it matters. Verifier defaults to FAIL. Prod-Readiness fires automatically on sensitive:* plans (auth, payments, PII, network, reliability).
  • Tool-neutral. Role contracts live in .agents/ — the same pipeline runs in Claude Code slash commands, GitHub Copilot paste workflows, or any AGENTS.md-aware tool.

Quick start

# Once per machine
claude plugin marketplace add rgadda/frontend-eng-team --scope user
claude plugin install frontend-team@frontend-team-marketplace --scope user

# Once per target repo
cd /path/to/your-frontend-project
ft-bootstrap

# Verify
/agents    # lists five subagents
/pipeline  # runs the pipeline

Full setup + troubleshooting: scripts/setup-global.md.

What's inside

Layer Items
Roles (.agents/roles/) PM · Architect · Implementer · Reviewer · Prod-Readiness · Verifier
Orchestrators (.agents/) pipeline.md (full) · pipeline-lite.md (small, non-sensitive changes)
Subagents (plugins/frontend-team/agents/) repo-explorer · frontend-reviewer · test-runner · prod-readiness · verifier
Skills (plugins/frontend-team/skills/) component-conventions · pr-prep · pm-clarify · loop-engineering
Metrics (scripts/) collect-pr-metrics.ts — before/after PR velocity report · claude-usage-report.py — local Claude Code usage + leak-signal report

Usage

Run the full pipeline on a new feature:

/pipeline Add a UserProfile component to the settings feature
          that fetches user data via Axios and displays it
          with editable fields and form validation

The Architect plans, you approve, then Implementer → Reviewer → Verifier run through. One slash command, one approval gate, one PR-ready output.

For small changes:

/pipeline-lite Add a loading spinner to SettingsPanel while user data is fetching

Lite mode skips PM, Architect, and Verifier. Target cost ≤25K tokens (~55% cheaper than full). A hard entry gate refuses lite mode on any task that mentions auth / payments / PII / tokens / cookies — those always run through /pipeline.

When to use which:

Situation Use
New feature with user-facing behavior /pipeline
Sensitive change (auth, payments, PII, security, reliability) /pipeline — always
Refactor touching multiple modules or a Context / hook / effect / portal /pipeline
Change > 200 LOC or > 3 files /pipeline
Small bugfix, config tweak, copy change, one-file typo /pipeline-lite
CSS-only change, docs update /pipeline-lite

Works with

  • Claude Code — slash commands in .claude/commands/ (/pm, /architect, /implement, /review, /verify, /pipeline, /pipeline-lite).
  • GitHub Copilot in VS Code — paste-driven workflow, see .github/copilot-instructions.md.
  • Any AGENTS.md-aware tool — root AGENTS.md routes into .agents/roles/ and .agents/pipeline.md.

Tech stack this template is tuned for

React 18 · TypeScript 5 (strict) · Vite · CSS Modules · Axios · Jest + RTL · Playwright · ESLint + Prettier.

Stack-specific conventions live in .agents/conventions.md. To adapt the template, edit that file plus CLAUDE.md; the role contracts themselves are stack-agnostic.

Docs

License

MIT — see LICENSE.

Yorumlar (0)

Sonuc bulunamadi