Skip to content

damusix/ai-tools

Repository files navigation

claude-marketplace

A monorepo for Claude Code plugins and standalone AI tooling. Install the marketplace for two practical plugins (persistent memory and safer Bash auto-approval), or use the standalone ralph-loop for autonomous multi-iteration coding.

Quickstart

Add this marketplace, then install both plugins:

/plugin marketplace add damusix/ai-tools
/plugin install ai-memory@damusix-ai-tools
/plugin install auto-approve-compound-bash@damusix-ai-tools

What you get

ai-memory

Give Claude long-term project memory across sessions, with local-first storage and tools to organize context.

  • Captures observations and synthesizes reusable memories
  • Shares one memory service across Claude Code sessions
  • Provides MCP tools for saving, searching, and organizing memory
  • Includes dashboard UI to browse and manage memories
  • Adds /remember and /forget slash commands

Docs: ai-memory/README.md

auto-approve-compound-bash (cc-auto-approve-fix source)

Auto-approve compound Bash commands safely by parsing each command segment against allow/deny rules.

  • Uses a native Go shell parser (mvdan.cc/sh/v3/syntax) for AST-based checks
  • Handles compound operators, substitutions, subshells, and nested bash/sh/zsh -c
  • Falls through safely when uncertainty is detected
  • Supports --explain, simulate, and doctor workflows
  • Ships prebuilt binaries for darwin/linux and amd64/arm64

Docs: cc-auto-approve-fix/README.md

Standalone tools

ralph-loop

An autonomous coding loop that drives AI agents (Claude, Amp, Codex, OpenCode) through iterative development cycles. Not a Claude Code plugin — ralph is a standalone zx script that runs inside a self-contained Docker environment.

  • Splits large tasks into focused iterations — one job, one commit, one status update per cycle
  • Each iteration starts with a fresh agent session, keeping context clean and decisions sharp
  • Quality gates stash broken work and log failures so the next iteration can recover
  • Streams real-time output (text, tool calls, timestamps) while the agent works
  • Includes a Dockerfile with all runtimes and tooling pre-installed — nothing to configure on the host
docker compose up -d --build
docker exec -it ralph-wiggum zsh
cd ~/my-project && ralph init

Docs: ralph-loop/README.md

Why this marketplace

  • One setup path gives you both plugins immediately
  • Plugin names are stable in the damusix-ai-tools marketplace catalog
  • Memory stays local and command auto-approval remains rule-driven

Contributing

This repo is a pnpm workspace. Install dependencies from the root:

pnpm install

Versioning

Plugin versions are managed by a CLI tool at scripts/cli/. It bumps versions, syncs all manifest files, generates changelogs from git history, and creates git tags.

Interactive mode — select plugins and bump types via prompts:

pnpm dev version

Flag mode — bump a single plugin directly:

pnpm dev version ai-memory patch
pnpm dev version auto-approve-compound-bash minor

Plugin names come from .claude-plugin/marketplace.json, not directory names. The CLI requires a clean working tree before running.

What it does per plugin:

  1. Bumps the version in the source-of-truth file (package.json or plugin.json)
  2. Syncs version across marketplace.json and plugin.json
  3. Regenerates CHANGELOG.md from conventional commits (feat, fix, breaking changes)
  4. Commits as release: <plugin>@<version> and creates a git tag

Adding a new plugin

  1. Add the plugin directory to pnpm-workspace.yaml
  2. Add an entry to .claude-plugin/marketplace.json with name, version, and source
  3. The versioning CLI discovers it automatically

License

MIT