Scaffold new projects or retrofit existing ones with standard dev configs (linters, rules, hooks, testing)
Claude Code plugins and sandboxed Docker environments for AI coding assistants.
Install via the Claude Code marketplace:
/plugin marketplace add ShravanSunder/ai-tools
Then install individual plugins with /plugin install <name>@ai-tools.
| Plugin | Description |
|---|---|
ai-scaffold | Project scaffolding with standard dev configs (biome, ruff, vitest, pytest, cursor rules, claude hooks) |
skill-peekaboo | Visual UI testing for macOS apps using Peekaboo CLI |
quorum-counsel | Multi-model review orchestration -- counsel-reviewer and codex-solver background agents |
See plugins/ for full details.
Run AI coding agents (Claude Code, Codex, Gemini CLI) inside Docker containers with network-level isolation. The agent gets full workspace access but can only reach domains you explicitly allow.
# From any git repository
run-agent-sidecar.sh --run-claude
AI agents with tool use can execute arbitrary shell commands, install packages, and make network requests. Running them in a container with an egress firewall provides a practical security boundary without giving up functionality:
flowchart LR
subgraph host ["Host Machine"]
ctl["sidecar-ctl.sh\n(firewall control)"]
run["run-agent-sidecar.sh"]
end
subgraph container ["Docker Container"]
fw["iptables + dnsmasq\n(egress firewall)"]
agent["AI Agent\n(claude / codex / gemini)"]
ws["/workspace\n(bind mount)"]
end
run -->|"build + start"| container
ctl -->|"allow / block domains"| fw
agent -->|"read/write files"| ws
agent -->|"network requests"| fw
fw -->|"allowed domains only"| internet["npm, pypi, AI APIs\n(allowlisted)"]
fw -.->|"blocked"| blocked["everything else"]
| Concern | How it's handled |
|---|---|
| Arbitrary network access | Egress firewall blocks all traffic except allowlisted domains (npm, pypi, AI APIs). Toggle presets for GitHub push, Notion, Linear. |
| Persistent state leaking | Named volumes isolate shell history, venvs, node_modules per workspace. Container recreation is cheap (--reload ~5s). |
| Package supply chain | APT repos blocked at firewall level after build. Runtime installs impossible. |
| Git corruption | .git/ mounted read-only. Agent can read history but cannot rewrite refs or force-push. |
| Config tampering | .agent_sidecar/ shadowed with empty tmpfs. Agent cannot read or modify sidecar configuration. |
| Scope creep across repos | Each repo gets its own container, volumes, and firewall rules. No cross-repo contamination. |
Three-tier config hierarchy lets you customize per-team and per-developer without forking:
Base (agent_sidecar/setup/) -- defaults shipped with this repo
+ Repo (.agent_sidecar/*.repo.*) -- team overrides, committed
+ Local (.agent_sidecar/*.local.*) -- personal overrides, gitignored
Additive files (firewall allowlists, zshrc, init scripts) merge all tiers. Override files (config, dockerfile) pick the highest-priority tier.
run-agent-sidecar.sh --run-claude # Start Claude Code in sidecar
run-agent-sidecar.sh --run-codex # Start Codex
run-agent-sidecar.sh --run-gemini # Start Gemini CLI
run-agent-sidecar.sh --reload # Recreate container (~5s)
run-agent-sidecar.sh --full-reset # Rebuild image + recreate (~2-5min)
sidecar-ctl firewall allow notion # Allow Notion API
sidecar-ctl firewall toggle 15m # Enable all presets for 15 minutes
sidecar-ctl firewall clear # Revoke all toggle access
sidecar-ctl status # Show container + firewall state
See agent_sidecar/ for full setup, architecture, and configuration docs.
ai-tools/
├── plugins/ # Claude Code plugins
│ ├── ai-scaffold/ # Project scaffolding
│ ├── skill-peekaboo/ # macOS visual UI testing
│ └── quorum-counsel/ # Multi-model review orchestration
├── skills/ # Pure skills (future)
├── agent_sidecar/ # Docker sidecar system
└── CLAUDE.md # Agent instructions
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.
npx claudepluginhub shravansunder/ai-tools --plugin ai-scaffoldMulti-model review and analysis — agents for Claude (codex-solver, gemini-solver, oracle-solver, counsel-reviewer) and skills for Codex (claude-solver, gemini-solver, oracle-solver, counsel-reviewer) with confidence scoring, security auditing, and CLI permission hooks
Scaffold new projects and add features with best-practice templates
Universal Claude Code workflow with specialized agents, skills, hooks, and output styles for any software project. Includes orchestrator, code-reviewer, debugger, docs-writer, security-auditor, refactorer, and test-architect agents.
Bootstrap - Developer onboarding accelerator. Analyzes project requirements, detects missing dependencies, generates Docker/docker-compose configs, creates .env templates, sets up pre-commit hooks, configures IDE settings (VSCode/Cursor), and troubleshoots 'it works on my machine' issues. Gets developers productive in minutes, not hours.
Makes a repo agent-ready: AGENTS.md, boundary tests, CI pipeline, GC scripts — based on OpenAI's harness engineering methodology
Analyze and enforce best practices for AI coding agent projects. Assess codebase readiness across 8 pillars with /readiness, then scaffold enforcement with /setup: TDD, secret scanning, file size limits, auto-generated docs, and git hooks.
11 agents, 35 skills, 18 commands, 9 hooks — spec-driven multi-agent orchestration for Claude Code, with optional cross-device semantic memory.