AtharvaSDeshpande

Fullstack Builder — Testing skill for Claude Code

Testing community

A multi-agent system orchestrated by a Product Manager agent.

How to install Fullstack Builder

This entry records only its repository, not the path inside it, so there is no exact command to give. Open AtharvaSDeshpande/fullstack-builder and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Fullstack Builder does

A multi-agent system orchestrated by a Product Manager agent. The PM interviews the client and captures requirements in one spec file. It then runs a team of specialist agents that work in parallel, message each other, and stay inside their scope through deterministic scripts not prompts. Build a site from scratch or fix and extend an existing one.

Alternatives in Testing

  • Gstack — 15,000+ Garry Tan's exact Claude Code setup: 6 opinionated tools that serve as CEO, Eng Manager, Release Manag 33.9k ★
  • Cc Sdd — Spec-driven development (SDD) for your team's workflow 3.1k ★
  • Regen API — Regenerate the OpenAPI spec and web API client after server API changes 2.5k ★

README

fullstack-builder

Spec-driven, multi-agent full-stack website builder and maintainer. Host-neutral: it is Markdown, Node scripts and JSON, so it is designed for any agent platform that can read a skill or instruction file and run shell commands. Only the Claude Code hook wiring is shipped and tested; for other platforms (for example Codex or Antigravity) you wire the hook yourself or run in advisory mode. How strongly scope is enforced depends on the host; see "Host support".

Files

bootstrap.md                   separate skill: interviews the client like a PM, writes site.spec.json
SKILL.md                       the orchestrator (CRISP, standing instruction, feedback loop, 14 agents plus a canary)
guard/guard.mjs                the hook that checks every tool call of the PM and every subagent
guard/adapters.json            maps a host's tool names and hook fields to the guard's canonical ones
guard/lib/adapter.mjs          applies that mapping
AGENTS.md                      one-paragraph pointer for platforms that read AGENTS.md
guard/lib/*.mjs                path scope, shell allowlist, spawn limits, run state
guard/profiles.json            the one place that says what each agent may write, run and use
scripts/setup.mjs              `init` (spec templates) and `install ` (hook, config, agent files)
scripts/guard-report.mjs       summary of blocks and spawns; `--canary` proves the hook is live
scripts/wave.mjs               plans parallel waves, audits and closes each wave
scripts/bus-report.mjs         shows agent-to-agent mail: open requests, blockers, unread
guard/lib/bus.mjs, waves.mjs   mail rules and wave state used by the hook
scripts/guard-selftest.mjs     replays 110 checks through the real hook, offline
scripts/validate-spec.mjs      validates a JSON spec (errors stop a run, warnings are logged)
scripts/new-run.sh             creates a unique run folder and run-log.md
scripts/close-run.sh           sets final status and adds a line to docs