Jenie AI tools

=============================
       _            _      
      | |          (_)     
      | | ___ _ __  _  ___ 
  _   | |/ _ \ '_ \| |/ _ \
 | |__| |  __/ | | | |  __/
  \____/ \___|_| |_|_|\___|

=============================

jenie is a proto skill library focused to work with small / medium development tasks, considering:

• Identify available tasks (user stories or bugs) and validate consistency using INVEST and 3C practices
• Implement validated tasks using TDD practices

Keep it simple, use jenie in any case you need to implement a small / medium feature or fix a bug

Run jenie with no arguments at any time to see the command list.

Available skills

• prepare-for-wishes — loads global guidelines, analyzes the project's tech stack and conventions, then asks what to do
• list-tasks — paginates available tasks with descriptions
• analyze-task — validates a task against INVEST + 3C, checks correctness, assesses doability
• implement-task — verifies repo + validation recency, creates a feature branch, runs TDD, implements, lints, commits

About Jenie's conscience

Like I said, jenie's main goal is to work with not huge development taks, but is designed to follow Senior developer practices based on:

• Precision — respond to exactly what's asked: one clarifying question when the request is ambiguous, no unsolicited alternatives, no unrequested features, no speculative execution.
• Efficiency — cap context usage: ls/grep before reading, skip files already in context, batch independent reads/writes/commits, keep output concise.
• Safety — verify before acting: never assume a file or dependency exists, cite architectural claims as file:line, ask explicit confirmation before destructive commands (rm -rf, force-push) or new dependencies, never commit secrets.
• Formatting — no preamble or filler, structured markdown output, every response tagged with [Active Agent: Role | Task: #ID].

When these conflict, priority is safety > precision > formatting > efficiency.

Supported languages

prepare-for-wishes recognizes a project's tech stack by looking for these build/manifest files:

• JavaScript / TypeScript — package.json
• Rust — Cargo.toml
• Python — pyproject.toml
• Ruby — Gemfile
• Go — go.mod
• C / C++ — CMakeLists.txt

Any other stack still gets a basic analysis, but without dedicated framework/tooling detection.

If the developer already has codebase-memory-mcp configured as an MCP server, prepare-for-wishes prefers it automatically for a faster, richer analysis (158 languages, call graphs, hotspots) — this is optional and not a new requirement; the manifest-based detection above is the fallback when it isn't present.

Requirements

• Node.js >= 18
• OpenCode and/or Claude Code — the Claude Code installer shells out to the claude CLI, so it must be on your PATH
• Local folder for the tasks to be implemented, by default, jenie will look for the docs folder

Installation

npm install -g @my-jenie/ai-tools

Project structure

<project_root>/
├── docs/
│   ├── user-stories/           # user story task files
│   └── bugs/                   # bug task files
└── .jenie/
    ├── technical-context.md        # generated by prepare-for-wishes
    └── .validation-registry.json   # analyze-task validation results

OpenCode

cd my-project
jenie install --opencode    # repository level (current directory)

OpenCode installs are always repository level — there's no --global option, since OpenCode looks for a config file per project.

The installer:

1. Finds the nearest opencode.json (or .jsonc), walking up from the current directory — creates one when none exists, and converts an existing .jsonc to .json.
2. Registers the plugin path in the config's plugin array (no-op if already present).

Restart OpenCode afterwards — plugins are loaded at startup.

How the commands appear: /jenie (entry point, defaults to prepare-for-wishes), /jenie prepare-for-wishes, /jenie list-tasks, /jenie analyze-task, /jenie implement-task.

Claude Code

cd my-project
jenie install --claude-code             # repository level (current directory)
jenie install --claude-code --global    # global (user scope, all projects)

Without --global, the installer works at repository level: it writes extraKnownMarketplaces and enabledPlugins into ./.claude/settings.json. Commit that file and teammates get prompted to install the plugin automatically when they open the project. With --global, the plugin is installed at user scope for all projects and the current directory doesn't matter.

The installer:

1. Ensures .claude-plugin/plugin.json and .claude-plugin/marketplace.json exist in the package — created if missing; an existing marketplace.json gets the jenie entry merged in without touching other entries.
2. Registers the package directory as a plugin marketplace and installs jenie@ai-tools through the claude CLI.
3. Removes any legacy .claude/skills/jenie install left by older versions (only after the plugin install succeeds).

Restart Claude Code afterwards (exit and run claude again) — plugins are discovered at startup.

How the commands appear: the command picker lists skills by their short name with the plugin as a label — /prepare-for-wishes (jenie), /list-tasks (jenie), /analyze-task (jenie), /implement-task (jenie). The namespaced form /jenie:<skill-name> also works when typed, but Claude Code only displays it when another plugin defines a skill with the same name.

Alternatively, without npm or the jenie CLI (requires the .claude-plugin/ manifests to be committed to the repo):

claude plugin marketplace add jersson/jenie-ai-tools
claude plugin install jenie@ai-tools

Uninstall

jenie uninstall --opencode                # repository level (current directory)
jenie uninstall --claude-code             # repository level (current directory)
jenie uninstall --claude-code --global    # global (user scope)

The Claude Code variant uninstalls the plugin, removes the marketplace registration, and cleans up legacy installs.

Version

jenie --version    # or jenie -v