C3, short for Christophe Agentic Collective, is a personal collective of agents and skills designed for agentic coding and various other purposes. It originated as Christophe’s Coding Crew, which is how the name C3 came to be. Over time, it has undergone significant evolution and has now become known as Christophe’s Agentic Collective. The C3 name has remained consistent throughout its journey. 😉
Caution
Before installing any plugin: Plugins can execute arbitrary commands on your machine. Always review a plugin's code before installing it.
Warning
This is my personal collective. It is in constant flux. I try to keep the plugin version stable and usable, but YMMV 😇
My agentic workflow is built on a simple belief: create small automation steps, use them, and iteratively improve. Each skill and agent emerged from real needs, was refined through use, and continues to evolve.
flowchart LR
A[Idea] --> B(["/develop-skill"])
A --> C(["/develop-agent"])
B --> D[skills/]
C --> E[agents/]
D --> F["make local"]
E --> F
F --> G[Use in Sessions]
G --> H(["/lessons-learned"])
H --> I[Auto-Improve]
I --> J{Stable?}
J -->|No| G
J -->|Yes| K[Push to GitHub]
- Create: Use
/develop-skillor/develop-agentto design new skills/agents - Test locally:
make localruns with--plugin-dir ./ - Use/Refine loop: Use in sessions, run
/lessons-learnedto capture improvements - Distribute: Push to GitHub when stable
Install from the christophe.vg marketplace:
# Add the marketplace
claude plugin marketplace add christophevg/marketplace
# Install C3
claude plugin install c3@christophe.vgSkills and agents are namespaced (e.g., /c3:python, /c3:commit).
To develop or test the latest version locally:
# Clone the repository
git clone https://github.com/christophevg/c3.git
cd c3
make installNote
make install copies Makefile.clause to your personal ~/.claude folder and allows your project Makefiles to include it using -include ~/.claude/Makefile. Now you can use several utility targets to work with claude and use C3 as a harness configuration. For example make basically runs Claude Code using Ollama. If you do this from the C3 folder, it will also include that folder as a plugin.
Skills provide focused guidance for specific technologies and workflows.
| Skill | Description |
|---|---|
/mcp-server |
Guide for designing and building MCP servers (FastMCP, security, deployment). |
/plugin-development |
Guide for creating Claude Code plugins (structure, manifest, distribution). |
| Skill | Description |
|---|---|
/project |
Dispatcher for project management skills. |
/project-feature |
Capture and scope new features. |
/project-manage |
Full implementation workflow with specialized agents. |
/project-status |
Generate STATUS.md with executive summary, metrics, dependencies, blockers, risks. |
/project-todo-refine |
Iteratively refine TODO.md topics by reviewing state, scope, and priority. |
/website-manage |
Manage content websites with conversational workflow. No PRs, no agents. |
| Skill | Description |
|---|---|
/pa |
Main dispatcher for personal assistant workflow. |
/pa-inbox |
Process inbox files into actionable TODOs. |
/pa-session |
Manage session state for workflow continuity. |
/pa-outbox |
Generate formatted replies and manage archive. |
| Skill | Description |
|---|---|
/python |
Python coding standards and testing patterns. |
/pymongo |
MongoDB/PyMongo patterns and security. |
/baseweb |
Baseweb/Vue/Vuetify best practices. |
/fire |
Python Fire CLI patterns. |
/textual |
Textual TUI framework. |
/rich |
Rich console output. |
/vuetify-v1 |
Vuetify 1.5 components in legacy Baseweb projects. |
/vuetify-v2 |
Vuetify V2 components in Baseweb projects. |
/vuetify-v3 |
Vuetify V3 components with comprehensive patterns and migration guide from V2. |
| Skill | Description |
|---|---|
/develop-skill |
Create and refine Claude Code skills. |
/develop-agent |
Develop Claude Code agents. |
| Skill | Description |
|---|---|
/commit |
Git commits with atomic commits and conventional format. |
/bug-fixing |
Systematic bug fixing with TDD. |
/git-activity-report |
Human-readable git activity summaries. |
/git-scripting |
Safe git command usage in scripts. |
/naming |
Choose names for projects, products, agents. |
/analysis-integration |
Integrate findings from multiple agents. |
/lessons-learned |
Review session to improve skills/agents. |
/documentation |
Sphinx/readthedocs setup. |
/markdown-to-pdf |
Convert Markdown to PDF with TOC. |
/readme |
Create and maintain README files. |
/transcribe-session |
Curated session transcripts. |
/api2mod |
Convert API docs to Python modules. |
/spec2mod |
Generate Python from OpenAPI specs. |
/start-baseweb-project |
Start new Baseweb projects. |
/vue-form-generator |
Schema-based Vue.js forms. |
/ollama |
Python ollama library for LLM integration. |
/pyenv |
Manage Python versions. |
/pypi-publish |
Publish packages to PyPI. |
/mcp-tools |
Work with MCP tools: discovery, naming, sub-agent config. |
| Agent | Description |
|---|---|
assistant |
Personal assistant for inbox processing. |
project-manager |
Project workflow orchestration with multi-task execution. |
git-manager |
Git operations via c3:commit skill. |
functional-analyst |
Requirements extraction and task planning. |
researcher |
Comprehensive research with provenance. |
api-architect |
API design and architecture. |
ui-ux-designer |
User experience and interface design. |
python-developer |
Python implementation. |
code-reviewer |
Code quality review. |
testing-engineer |
Test planning and coverage. |
security-engineer |
Security vulnerability assessment. |
end-user-documenter |
Documentation generation. |
knowledge-agent |
Knowledge base querying and evolution. |
The plugin includes MCP servers that provide tools for Claude Code.
| Server | Tools | Description |
|---|---|---|
email |
9 tools | Email exchange via IMAP/SMTP with security hardening |
Tools: list_accounts, list_folders, search_emails, get_email, download_attachment, send_email, reply_email, move_email, delete_email
Prerequisites: uv must be installed. The server uses uv run to automatically manage dependencies.
Configuration via environment variables:
export EMAIL_IMAP_HOST=imap.gmail.com
export EMAIL_SMTP_HOST=smtp.gmail.com
export EMAIL_USERNAME=your-email@gmail.com
export EMAIL_PASSWORD=your-app-passwordOptional recipient whitelist:
export EMAIL_RECIPIENT_DOMAINS=example.com,company.orgThe /project skill dispatcher handles one-off operations, while the project-manager agent orchestrates multi-task sessions with progress tracking and memory persistence.
flowchart TB
subgraph Phase0["Phase 0: Project State"]
A[Start] --> B{State?}
B -->|New Project| C[Phase 1A: Initial Analysis]
B -->|Incomplete Setup| D[Phase 1B: Review & Backlog]
B -->|Ready for Work| E[Check Unsorted Items]
E --> F{Unsorted?}
F -->|Yes| G[Sort or Skip?]
F -->|No| H[Propose Next Task]
end
subgraph Phase1["Phase 1: Analysis"]
C --> I[Functional Analyst]
D --> I
I -->|Optional| J[Researcher]
J -->|Tech recommendations| I
I -->|functional.md, TODO.md| K[Task Scope Classification]
G -->|Sort| I
G -->|Skip| H
end
subgraph Phase2["Phase 2: Domain Review"]
K --> L{Scope?}
L -->|Backend| M[API Architect]
L -->|Frontend| N[UI/UX Designer]
L -->|Full Stack| M
L -->|Full Stack| N
L -->|Security-related| O[Security Engineer]
M --> P[Phase 3: Consensus]
N --> P
O --> P
end
subgraph Phase4["Phase 4: Implementation (per task)"]
H --> Q[Plan]
Q --> R[Python Developer]
R --> S{Review Cycle}
S -->|Functional Review| T[Functional Analyst]
T -->|Pass| U{Domain Reviews}
U -->|Parallel| M2[API Architect]
U -->|Parallel| N2[UI/UX Designer]
U -->|Parallel| O2[Security Engineer]
M2 --> V{Quality Reviews}
N2 --> V
O2 --> V
V -->|Parallel| W[Code Reviewer]
V -->|Parallel| X[Testing Engineer]
W --> Y{User-facing?}
X --> Y
Y -->|Yes| Z[End-User Documenter]
Y -->|No| AA{All Approve?}
Z --> AA
AA -->|No| R
AA -->|Yes| AB[Complete Task]
AB --> AC{More Tasks?}
end
P --> Q
AC -->|Yes| Q
AC -->|No| AD[Project Complete]
See CONTRIBUTING.md for guidelines.