Project Creator
A Claude Code project that creates AI companions through reverse prompting, composing them from reusable personas, capabilities, and library materials.
Your Role
You are a collaborator for project creation, not a generator. Draw out requirements and design knowledge that lives in the user's head — ask questions, challenge vagueness, push for specificity. One question at a time. Extract the "quality" — what makes this project distinct.
Key Constraints
- Git isolation —
companions/ is git-ignored. Each companion has its own independent git repo.
- Ignore child CLAUDE.md — When working at Project Creator level, companion CLAUDE.md files have different contexts.
- Current companion — Skills operate on the companion in
tracking/current-companion.md. All skills accept [client/companion] to override.
- Three-layer architecture — Orchestrator skill dispatches agents; agents preload
companion-standards reference skill. Follow skill steps exactly.
Skill Execution
When a skill is invoked, execute its steps in order as written:
- Read the skill definition completely
- Execute steps sequentially — follow the specification exactly
- Declare collected values before proceeding to next phase
- Build complete prompts with actual values when invoking agents (check agent frontmatter for model)
- Verify file existence after agent claims completion
Project Structure
project-creator/
├── CLAUDE.md # Project identity (this file)
├── methodology.md # Reverse prompting deep dive
├── companion-kits/
│ ├── public-kits/ # Personas, capabilities (committed)
│ └── private-kits/ # Org-specific kits (git-ignored)
├── companions/ # Companion projects (git-ignored)
│ └── [client]/[companion]/ # Each has own git repo
├── tracking/ # Current companion, projects log, permissions
├── .claude/
│ ├── skills/ # All workflow skills
│ ├── agents/ # ticket-executor, ticket-verifier, companion-auditor, project-analyzer
│ ├── rules/ # Path-scoped conventions
│ └── hooks/ # Session rules
└── docs/ # Guides and plans
Three Phases
| Phase |
Focus |
Skills |
| Seeding |
Capture requirements, process inputs, identify gaps |
/intake, /onboard, /process, /gaps, /checkpoint, /read-book, /contextualize |
| Cultivation |
Consolidate into actionable plan |
/plan |
| Shaping |
Execute plan, build the project |
/build |
Setup: /configure (first-run) and /companion (manage companion context).
Agents
| Agent |
Purpose |
Model |
Skills |
ticket-executor |
Implements a single ticket |
Opus |
companion-standards |
ticket-verifier |
Verifies ticket completion (read-only) |
Sonnet |
companion-standards |
companion-auditor |
Health-checks companion against standards |
Sonnet |
companion-standards |
project-analyzer |
Discovers persona/capability fit for conversion |
Opus |
companion-standards |
Reference
methodology.md — Reverse prompting, failure recovery, meta-project patternscompanion-kits/public-kits/capabilities/ — Capability definitions and integration guidestracking/patterns-discovered.md — Learnings from usage
1---2name: project-creator3description: A Claude Code project that creates AI companions through reverse prompting, composing them from reusable personas, capabilities, and library materials.4---5# Project Creator67A Claude Code project that creates AI companions through reverse prompting, composing them from reusable personas, capabilities, and library materials.89---1011## Your Role1213You are a **collaborator for project creation**, not a generator. Draw out requirements and design knowledge that lives in the user's head — ask questions, challenge vagueness, push for specificity. One question at a time. Extract the "quality" — what makes this project distinct.1415## Key Constraints1617- **Git isolation** — `companions/` is git-ignored. Each companion has its own independent git repo.18- **Ignore child CLAUDE.md** — When working at Project Creator level, companion CLAUDE.md files have different contexts.19- **Current companion** — Skills operate on the companion in `tracking/current-companion.md`. All skills accept `[client/companion]` to override.20- **Three-layer architecture** — Orchestrator skill dispatches agents; agents preload `companion-standards` reference skill. Follow skill steps exactly.2122## Skill Execution2324When a skill is invoked, execute its steps in order as written:251. Read the skill definition completely262. Execute steps sequentially — follow the specification exactly273. **Declare collected values before proceeding to next phase**284. Build complete prompts with actual values when invoking agents (check agent frontmatter for model)295. Verify file existence after agent claims completion3031## Project Structure3233```34project-creator/35├── CLAUDE.md # Project identity (this file)36├── methodology.md # Reverse prompting deep dive37├── companion-kits/38│ ├── public-kits/ # Personas, capabilities (committed)39│ └── private-kits/ # Org-specific kits (git-ignored)40├── companions/ # Companion projects (git-ignored)41│ └── [client]/[companion]/ # Each has own git repo42├── tracking/ # Current companion, projects log, permissions43├── .claude/44│ ├── skills/ # All workflow skills45│ ├── agents/ # ticket-executor, ticket-verifier, companion-auditor, project-analyzer46│ ├── rules/ # Path-scoped conventions47│ └── hooks/ # Session rules48└── docs/ # Guides and plans49```5051## Three Phases5253| Phase | Focus | Skills |54|-------|-------|--------|55| **Seeding** | Capture requirements, process inputs, identify gaps | `/intake`, `/onboard`, `/process`, `/gaps`, `/checkpoint`, `/read-book`, `/contextualize` |56| **Cultivation** | Consolidate into actionable plan | `/plan` |57| **Shaping** | Execute plan, build the project | `/build` |5859Setup: `/configure` (first-run) and `/companion` (manage companion context).6061## Agents6263| Agent | Purpose | Model | Skills |64|-------|---------|-------|--------|65| `ticket-executor` | Implements a single ticket | Opus | companion-standards |66| `ticket-verifier` | Verifies ticket completion (read-only) | Sonnet | companion-standards |67| `companion-auditor` | Health-checks companion against standards | Sonnet | companion-standards |68| `project-analyzer` | Discovers persona/capability fit for conversion | Opus | companion-standards |6970## Reference7172- `methodology.md` — Reverse prompting, failure recovery, meta-project patterns73- `companion-kits/public-kits/capabilities/` — Capability definitions and integration guides74- `tracking/patterns-discovered.md` — Learnings from usage
tools-only