project-finisher

project-finisher

Ever wish you could hand Claude Code a goal file and walk away while it figures out the architecture, writes the code, tests everything, and keeps going until it's done? That's what this plugin does. You describe what you want, point it at a project, and it runs an iterative loop — brainstorming approaches, planning implementation, writing code, and reviewing its own work — until the goal is met. It remembers where it left off between sessions, so even large projects can be tackled across multiple sittings.

Workflow Overview

flowchart TD
    START(["/finish --goal goal.md"]) --> CHECK{project_memory/ exists?}
    CHECK -- No --> INIT["Initialize memory Propose 1-3 milestones"]
    CHECK -- Yes --> RESUME["Resume from current phase"]
    INIT --> BRAINSTORM
    RESUME --> BRAINSTORM

    subgraph BRAINSTORM ["Phase 1: Brainstorm"]
        direction TB
        B1["Round 1: Feasibility & approach /scientific-brainstorming"]
        B1 -- "pushback" --> B2["Round 2: Address concerns /scientific-brainstorming"]
        B2 -- "new concerns?" --> B3["Round 3+: Until convergence /scientific-brainstorming"]
        B3 --> BOUT["Re-scope milestone Revise roadmap Record decisions"]
    end

    BRAINSTORM --> PLAN

    subgraph PLAN ["Phase 2: Plan"]
        P1["Create step-by-step plan Files, tasks, tests, dependencies Save to docs/plans/"]
    end

    PLAN --> EXECUTE

    subgraph EXECUTE ["Phase 3: Execute"]
        E1["Implement plan steps Write code + tests Incremental commits on pf/milestone-N Handle blockers"]
    end

    EXECUTE --> REVIEW

    subgraph REVIEW ["Phase 4: Review"]
        R1["Score each quality dimension (1-10) Verify acceptance criteria Run test suite Check regressions Update lessons.md Doc-check + update docs Squash-merge to default branch Archive milestone branch"]
    end

    REVIEW --> DONE{Goal reached?}
    DONE -- "Yes" --> STOP(["Done!"])
    DONE -- "No" --> NEXT["Next milestone"] --> BRAINSTORM

    style BRAINSTORM fill:#e8f4fd,stroke:#4a90d9
    style PLAN fill:#e8fde8,stroke:#4a9950
    style EXECUTE fill:#fdf4e8,stroke:#d9904a
    style REVIEW fill:#f4e8fd,stroke:#904ad9
    style START fill:#333,color:#fff,stroke:#333
    style STOP fill:#333,color:#fff,stroke:#333

Roadmap changes happen at two points: Brainstorm (re-scope, add prerequisites, split, reorder) and Review (propose new milestones, re-prioritize).

The plugin stops to ask you when the goal is ambiguous, approaches are equally viable, external resources are needed, the project has diverged from the goal, or the same milestone has failed twice. In --auto mode, these are resolved autonomously (except external resources).

Prerequisites

This plugin depends on skills from these plugins:

Plugin	Used In	Purpose
scientific-skills@claude-scientific-skills	Brainstorm phase	/scientific-brainstorming for feasibility analysis
superpowers@claude-plugins-official	Plan phase	/superpowers:write-plan for structured implementation plans

Install them first if you don't already have them:

claude plugin marketplace add https://github.com/K-Dense-AI/claude-scientific-skills.git
claude plugin install scientific-skills@claude-scientific-skills

claude plugin install superpowers@claude-plugins-official

Installation

# 1. Register the marketplace
claude plugin marketplace add https://github.com/yuanhao96/project-finisher.git

# 2. Install the plugin
claude plugin install project-finisher@project-finisher

# 3. Restart Claude Code to activate

Usage

Starting a Project

Point the plugin at a goal file that describes what you want to build or finish:

/finish --goal path/to/goal.md

To target a specific project directory (defaults to the current working directory):

/finish --goal path/to/goal.md --project ./my-project

The plugin will read your goal, assess the current state of the project, and begin iterating autonomously until the goal is satisfied.

Auto Mode

Add --auto to run with minimal user interaction. The plugin will make decisions autonomously instead of stopping to ask:

/finish --goal path/to/goal.md --auto

In auto mode:

• Initial milestones are approved automatically
• Brainstorming choices (approach selection, trade-off decisions) are resolved using the recommended or simplest option
• Blocked milestones are re-scoped or skipped after repeated failures
• The only exception is external resources (API keys, credentials) — the plugin still stops for those

All autonomous decisions are logged with an [AUTO] prefix in project_memory/current_context.md so you can review them after the session.

Quality Scoring

Milestones are evaluated on multiple weighted dimensions (1-10 per dimension) instead of just binary pass/fail. Add a ## Quality Priorities section to your goal file to customize: