valzav

Valcraft — Testing skill for Claude Code

Testing community

Agent skills for spec-driven delivery — one plugin for Claude Code, OpenAI Codex, and OpenCode.

How to install Valcraft

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

What Valcraft does

Agent skills for spec-driven delivery — one plugin for Claude Code, OpenAI Codex, and OpenCode. At the center: an agentic delivery loop (plan → review → implement → review → merge) over fresh-context worker agents; around it, cast scaffolds specs in the repo, spec turns PRDs into feature specs, temper learns from what shipped.

Alternatives in Testing

Spec Kitty — Spec-Driven Development for serious software developers 1.6k ★
Spec Driven Develop — Spec-driven development workflow for AI coding agents: architecture-first planning, task decomposition, GitHub 975 ★
Spectra App — Spec-driven development for coding agents — a desktop app, a CLI, and skills for Codex, Claude Code, Cursor, C 735 ★

README

valcraft

Agent skills for spec-driven delivery, packaged as one plugin for Claude Code, OpenAI Codex, OpenCode, and Cursor (Teams or Enterprise with marketplace-import authority). At the center is an agentic **delivery loop** — draft → review → forge → review → land — run over fresh-context worker agents, inside one Claude Code, Codex, or Cursor session or through an orchestrator over several instances. Around it: `cast` creates the project frame, `spec` creates every feature or quick contract, and `temper` learns from what shipped.

Status: alpha.

Problems it addresses

If you have seen this…	valcraft's answer
The agent forgets requirements between sessions and reinvents them.	`cast` establishes the project frame and `spec` writes git-owned contracts with stable IDs (`FR-`, `AC-`, `T-`, `ADR-`) that plans, commits, tests, and reviews cite. Context lives with the code.
"Make X" turns into a pile of unreviewed code.	`foreman` coordinates independent plan and code reviews; `land` finalizes only the exact reviewed target, so implementer verification never becomes approval.
One long session runs out of context or reports work it never did.	`foreman` keeps its own context small — every worker starts cold, reports land on disk, and a run resumes from the tracker, git, and those reports.
Either you approve every step, or the agent runs away.