romerma/mstack
Rigorous, verifiable agent workflows for Claude Code: gates that are code, not prose
field: unknown workflow: PR / code reviewPlanning & specsGit & commitsTesting & TDDFrontend & design
ARCHETYPE
Orchestrator
A bench of specialised subagents. The main agent mostly delegates.
Copy this rig
# review before running: this installs third-party code $ npx degit romerma/mstack/agents ./rig-mstack/agents $ npx degit romerma/mstack/skills ./rig-mstack/skills $ npx degit romerma/mstack/hooks ./rig-mstack/hooks
MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.
$ npx degit romerma/mstack/skills/commit .claude/skills/commit $ npx degit romerma/mstack/skills/design .claude/skills/design $ npx degit romerma/mstack/skills/implement .claude/skills/implement $ npx degit romerma/mstack/skills/orchestrate .claude/skills/orchestrate $ npx degit romerma/mstack/skills/reflect .claude/skills/reflect $ npx degit romerma/mstack/skills/review .claude/skills/review $ npx degit romerma/mstack/skills/router .claude/skills/router $ npx degit romerma/mstack/skills/setup .claude/skills/setup $ npx degit romerma/mstack/skills/ship .claude/skills/ship $ npx degit romerma/mstack/skills/spec .claude/skills/spec $ npx degit romerma/mstack/skills/test-first .claude/skills/test-first $ npx degit romerma/mstack/skills/understand .claude/skills/understand $ npx degit romerma/mstack/skills/unslop .claude/skills/unslop $ npx degit romerma/mstack/skills/verify .claude/skills/verify
$ curl -fsSL --create-dirs -o .claude/agents/implementer.md https://raw.githubusercontent.com/romerma/mstack/main/agents/implementer.md $ curl -fsSL --create-dirs -o .claude/agents/orchestrator.md https://raw.githubusercontent.com/romerma/mstack/main/agents/orchestrator.md $ curl -fsSL --create-dirs -o .claude/agents/reviewer.md https://raw.githubusercontent.com/romerma/mstack/main/agents/reviewer.md $ curl -fsSL --create-dirs -o .claude/agents/spec-author.md https://raw.githubusercontent.com/romerma/mstack/main/agents/spec-author.md $ curl -fsSL --create-dirs -o .claude/agents/spec-reviewer.md https://raw.githubusercontent.com/romerma/mstack/main/agents/spec-reviewer.md
This rig commits no guardrails. Here is the community baseline instead — the deny/ask rules most often found across all 16.2k rigs:
{
"permissions": {
"deny": [
"Read(./.env)",
"Read(./.env.*)",
"Read(**/.env)",
"Read(~/.ssh/**)",
"Bash(git push --force:*)",
"Read(**/*.pem)",
"Read(.env)",
"Read(~/.aws/**)",
"Read(**/*.key)",
"Read(**/.env.*)",
"Bash(git push --force*)",
"Bash(rm -rf:*)",
"Bash(rm -rf *)",
"Bash(sudo *)",
"Bash(sudo:*)",
"Bash(rm -rf /)",
"Bash(git push -f:*)",
"Read(.env.*)",
"Bash(git reset --hard*)",
"Bash(rm -rf /*)"
],
"ask": [
"Bash(git push:*)",
"Bash(git push *)",
"Bash(gh pr merge *)",
"Bash(git commit:*)",
"Bash(rm *)",
"Bash(gh pr merge:*)",
"Bash(gh pr create *)",
"Bash(git reset --hard:*)",
"Bash(npm publish:*)",
"Bash(git rebase *)"
]
}
} Skills (14)
Subagents (5)
| implementer model: inherit | Implements exactly one mstack work item, writes the tests that prove it, and records the requirement-to-test map. Never marks its own work approved or done. |
| orchestrator model: inherit | Owns one work item from intake to close. Selects the next role, enforces the lifecycle, and monitors every launched agent to a terminal result. Never writes application code and never approves its own |
| reviewer model: inherit | Independently reviews one mstack implementation against its requirements, its tests and the real diff. Runs the verification itself. Returns APPROVED or CHANGES_REQUESTED and never edits the code. |
| spec-author model: inherit | Writes the spec for one item under the mstack spec path. Produces proposal, design, tasks and requirements with stable R-ids, then hands off. Never reviews its own spec. |
| spec-reviewer model: inherit | Reviews one mstack spec independently and adversarially before any code is written. Returns APPROVED or CHANGES_REQUESTED. Never edits the spec and never reviews a spec it wrote. |
Similar rigs
AdamAkhlaq/claude-code-toolkit
A browsable catalog of Claude Code building blocks — CLAUDE.md, settings, hooks, agents, and doc-writing skills — to copy into your own project and make your own.
Skill Collector 2.5k tok · · Q21
S-jooyoung/oh-my-joy
AI says it's done. OMJ makes it prove it. One workflow for Claude Code and Codex: interview → plan with an independent critique → approve once → build, review, and verify with recorded evidence. Figma-to-code with real browser checks built
Skill Collector 4.3k tok · · Q8
Dcastroro/sdd-starter
Local-first Spec-Driven Development template and Claude Code plugin: traceable requirements, scoped implementation workflows and deterministic validation.
Pragmatist 591 tok · · Q3
iamvonpasion/hashb
Dev workflow toolkit for Claude Code — 14 skills that turn AI into a structured development partner. Spec → Design → Eng → TDD → Review → QA → Ship.
Skill Collector 2.5k tok · · Q0