By Zpankz
Grill-driven scaffolding that designs complete workflow systems from the ground up. Uses structured interrogation to compose CLAUDE.md, rules, skills, hooks, agents, tools, and workflows as primitives within a systems-thinking architecture.
Generate agent definitions native to the target tool's runtime — Claude Code .claude/agents/*.md, Antigravity/Gemini agents with kind field, Hermes plugin-provided agents via register(ctx), Codex agent definitions, or Pi agent configs. Detects the active tool and emits the correct format. Invoked by scaffold or standalone via /generate-agents.
Generate a project CLAUDE.md file — the shared memory layer of the workflow system. Produces minimal, high-signal instructions that every other primitive (rules, skills, hooks, agents, workflows) references. Invoked by scaffold or standalone via /generate-claudemd.
Generate hooks native to the target tool's runtime — shell hooks, Python register(ctx), TypeScript pi.on(), hookify rules, or quality tool integrations (ruff, pyright, eslint, pytest). Detects the active tool and emits the correct format. Invoked by scaffold or standalone via /generate-hooks.
Generate .claude/rules/ files — the constraint layer of the workflow system. Produces focused rule files that hooks enforce and agents follow. Scoped by topic and optionally by file path. Invoked by scaffold or standalone via /generate-rules.
Generate .claude/skills/ definitions — the capability layer of the workflow system. Identifies recurring workflows and creates SKILL.md files that agents execute and workflows compose. Invoked by scaffold or standalone via /generate-skills.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Claude Code plugin that designs and builds complete workflow systems from the ground up. Uses grill-with-docs interrogation to understand your project, then composes CLAUDE.md, rules, skills, hooks, agents, tools, and workflows as primitives within a systems-thinking architecture.
The workflow is the product. Everything else is a building block.
claude plugin install https://github.com/Zpankz/create-workflow
pi install https://github.com/Zpankz/create-workflow
gemini extensions install https://github.com/Zpankz/create-workflow
hermes plugins install https://github.com/Zpankz/create-workflow
Add the repo as a self-hosted marketplace, then install:
codex plugin marketplace add https://github.com/Zpankz/create-workflow
codex plugin add create-workflow@create-workflow-marketplace
| Tool | Method | Manifest | Skills | Hooks | Agents |
|---|---|---|---|---|---|
| Claude Code | claude plugin install | .claude-plugin/plugin.json | Yes | Generator | Generator |
| Pi | pi install | package.json ("pi" key) | Yes | Generator | Generator |
| Antigravity | gemini extensions install | plugin.json (root) | Yes | Generator | Generator |
| Hermes | hermes plugins install | plugin.yaml + __init__.py | Yes | Generator | Generator |
| Codex | codex plugin marketplace add | .codex-plugin/plugin.json | Yes | Generator | Generator |
Skills are bundled and work natively. Hooks and agents are generated by the generate-hooks and generate-agents skills — they emit native output for whichever tool is running.
The generate-hooks and generate-agents skills detect which tool is running and emit native output:
| Generator | Claude Code | Antigravity | Hermes | Codex | Pi |
|---|---|---|---|---|---|
| generate-hooks | settings.json | settings.json | Python ctx.register_hook() | hooks.json | TypeScript pi.on() |
| generate-agents | .claude/agents/*.md | agents/*.md (with kind) | Python ctx.register_agent() | .agents/*.md | pi/agents/*.md |
Semantic (natural language, interpreted by LLM)
./CLAUDE.md — generalised context
.claude/agents/*.md — specialised context
.claude/skills/*/SKILL.md — conditionals (invoked on match or command)
.claude/rules/*.md — constraints
Programmatic (deterministic, executed by runtime)
.claude/settings.json — constraints (hooks: PreToolUse, PostToolUse, Stop)
.mcp.json — functions (MCP server configurations)
.claude/workflows/*.js — orchestration (sandboxed JS VM, gated)
| Skill | Command | Purpose |
|---|---|---|
| scaffold | /scaffold | Full pipeline: scan, grill, generate, verify |
| grill | /grill | Standalone interrogation with domain docs |
| generate-claudemd | /generate-claudemd | Generate CLAUDE.md (shared context) |
| generate-rules | /generate-rules | Generate .claude/rules/ (constraints) |
| generate-skills | /generate-skills | Generate .claude/skills/ (capabilities) |
| generate-hooks | /generate-hooks | Generate hooks — native to active tool's runtime |
| generate-agents | /generate-agents | Generate agents — native to active tool's format |
| generate-tools | /generate-tools | Configure MCP servers and tool integrations |
| generate-workflows | /generate-workflows | Generate workflow skills (orchestration, --dual-track for .js) |
/scaffold
|- Phase 1: Scan — explore the repo, build a system map
|- Phase 2: Grill — structured interrogation, one question at a time
| |- Recommend answers based on scan findings
| |- Explore code before asking
| |- Persist decisions to .claude/scaffold-decisions.md inline
| |- Maintain CONTEXT.md glossary inline
| '- Create ADRs sparingly (hard to reverse + surprising + real trade-off)
|- Phase 3: Compose — map decisions to primitives, generate in dependency order
| |- Detect active tool — emit native formats
| |- Check for existing files — diff, don't overwrite
| |- CLAUDE.md (shared context — everything references this)
| |- Rules (constraints — agents and hooks enforce these)
| |- Skills (capabilities — agents use these)
| |- Hooks (reflexes — native to tool runtime)
| |- Agents (specialists — native to tool format)
| |- Tools (instruments — agents and skills use these)
| '- Workflows (orchestration — compose everything into pipelines)
'- Phase 4: Verify — consistency, completeness, coherence
Each generator works independently:
Self-referential self-improving agent framework for Claude Code. Inspired by Meta's HyperAgents paper — evolutionary optimization of skills, agents, hooks, and code via LLM-driven mutation and fitness-based selection.
Recursive context modularization, dependency analysis, and hyperedge bridge extraction
Emit and verify reproducible governance artifacts for local .index artifacts.
npx claudepluginhub zpankz/create-workflow --plugin create-workflowPersistent file-based planning for AI coding agents. Crash-proof markdown plans (task_plan.md, findings.md, progress.md) that survive context loss and /clear, with an opt-in completion gate and multi-agent shared state. Manus-style. Works with Claude Code, Codex CLI, Cursor, Kiro, OpenCode and 60+ agents via the SKILL.md standard. Includes Arabic, German, Spanish, and Chinese (Simplified and Traditional).
v9.44.1 — Patch release for Gemini environment/version detection and qwen auth gating. Run /octo:setup.
Claude harness - A harness for solo developers (Vibecoders) to handle full-cycle contract development.
Core skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques
Harness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.