gsd

Get Shit Done - Full project planning and execution workflow. Handles project initialization with deep context gathering, automated research, roadmap creation, phase planning, and execution with verification.

Security Scan

The skill's requests and instructions are coherent with a full-project planning/execution tool — it reads/writes project files, runs shell/git operations, spawns subagents, and performs web research — but it's powerful and you should review the file- and repo-modifying behaviors before use.

Purpose & Capability

The name/description (end-to-end project planning & execution) matches what the SKILL.md actually does: deep questioning, research, roadmap/phase management, plan execution, verification, debugging, design flows, and spawning helper subagents. The file reads/writes, git commits/tags, and web research calls align with that purpose.

Instruction Scope

The instructions contain many concrete shell steps that read and write project files (.planning/*, package.json, code files), spawn subagents (sessions_spawn / Task), run git operations (commit, tag), and can delete files (example: rm .planning/REQUIREMENTS.md). All of these are in-scope for a project execution tool, but they are high-impact actions: they can modify or remove repository files and persist debug sessions. Pay attention to automatic commit behavior (COMMIT_PLANNING_DOCS) and any actions that run without an explicit confirmation step.

Install Mechanism

Instruction-only skill with no install spec or downloaded binaries — nothing is written by an installer. All behavior is described in SKILL.md and operates on the user's repo and platform-provided facilities. This is the lowest install risk category.

Credentials

The skill requests no environment variables or external credentials in its metadata. It does rely on repository-local files (git, package manifests, project code) and platform tools (WebFetch/WebSearch, sessions_spawn, Context7 MCP). Those dependences are proportional to research, mapping, and codebase analysis duties. There are no unrelated credential requests in the skill metadata.

Persistence & Privilege

always:false (normal). The skill spawns autonomous subagents (sessions_spawn / Task) and often sets cleanup="keep", persisting debug/session files under .planning/debug. It also performs git commits and tags. Autonomous subagents with kept sessions increase blast radius if misused, but this behavior is coherent with a multi-agent project orchestration workflow.

Assessment

This skill appears internally consistent with a project-management/execution tool, but it performs high-impact repo operations and spawns autonomous subagents. Before installing or running it: 1) Review .planning/config.json (the COMMIT_PLANNING_DOCS flag) and consider disabling automatic commits until you're comfortable. 2) Run it first in a throwaway or non-production repository to observe behaviors (it will create .planning/, commit files, tag, and may rm files). 3) Be mindful of the skill's use of WebFetch/WebSearch and Context7 (these may send project context externally for research). 4) Note that debug and subagent sessions persist under .planning/debug (cleanup="keep"); if you don't want persistent subagents, avoid flows that spawn them or check platform controls for subagent lifecycle. 5) If you need stricter control, ensure the agent runs with limited filesystem permissions or in an isolated environment and review any commits/tags before pushing to remotes.

<objective> GSD (Get Shit Done) provides a complete workflow for taking projects from idea to execution through systematic planning, research, and phase-based development.

Full workflow port from Claude Code - Includes:

• Deep questioning and context gathering
• Automated domain research (4 parallel researchers)
• Requirements definition and scoping
• Roadmap creation with phase structure
• Phase planning with research and verification
• Wave-based parallel execution
• Goal-backward verification

This is the complete GSD system, not a simplified version. </objective>

<intake> What would you like to do?

Core workflow commands:

• new-project - Initialize a new project with deep context gathering, research, requirements, and roadmap
• plan-phase [N] - Create execution plans for a phase (with optional research)
• execute-phase [N] - Execute all plans in a phase with wave-based parallelization
• progress - Check project status and intelligently route to next action
• debug [issue] - Systematic debugging with persistent state across context resets
• quick - Execute ad-hoc tasks with GSD guarantees but skip optional agents
• discuss-phase [N] - Gather context through adaptive questioning before planning
• verify-work [N] - Validate built features through conversational UAT
• map-codebase - Analyze existing codebase for brownfield projects
• pause-work - Create handoff when pausing mid-phase
• resume-work - Resume from previous session with full context
• add-todo [desc] - Capture idea or task for later
• check-todos [area] - List and work on pending todos
• add-phase <desc> - Add phase to end of milestone
• insert-phase <after> <desc> - Insert urgent decimal phase
• remove-phase <N> - Remove future phase and renumber
• new-milestone [name] - Start new milestone cycle
• complete-milestone <ver> - Archive milestone and tag
• audit-milestone [ver] - Verify milestone completion
• settings - Configure workflow toggles and model profile

Flags:

• plan-phase [N] --research - Force re-research before planning
• plan-phase [N] --skip-research - Skip research, plan directly
• plan-phase [N] --gaps - Gap closure mode (after verification finds issues)
• plan-phase [N] --skip-verify - Skip plan verification loop
• execute-phase [N] --gaps-only - Execute only gap closure plans

Usage:

• /gsd new-project - Start a new project
• /gsd plan-phase 1 - Plan phase 1
• /gsd execute-phase 1 - Execute phase 1
• /gsd progress - Check where you are and what's next
• /gsd debug "button doesn't work" - Start debugging session
• /gsd quick - Quick ad-hoc task without full ceremony
• Or just tell me what you want and I'll guide you through GSD

What GSD does:

1. Deep questioning - Understand what you're building through conversation
2. Research - 4 parallel researchers investigate domain (stack, features, architecture, pitfalls)
3. Requirements - Define v1 scope through feature selection
4. Roadmap - Derive phases from requirements (not imposed structure)
5. Phase planning - Create executable plans with tasks, dependencies, verification
6. Execution - Run plans in parallel waves with per-task commits
7. Verification - Check must_haves against actual codebase </intake>

<routing> Based on user input, route to appropriate workflow:

Intent	Workflow
"new project", "initialize", "start project"	workflows/new-project.md
"new-project" (explicit)	workflows/new-project.md
"plan phase", "plan-phase", "create plan"	workflows/plan-phase.md
"execute phase", "execute-phase", "start work"	workflows/execute-phase.md
"progress", "status", "where am I"	workflows/progress.md
"debug", "investigate", "bug", "issue"	workflows/debug.md
"quick", "quick task", "ad-hoc"	workflows/quick.md
"discuss phase", "discuss-phase", "context"	workflows/discuss-phase.md
"verify", "verify-work", "UAT", "test"	workflows/verify-work.md
"map codebase", "map-codebase", "analyze code"	workflows/map-codebase.md
"pause", "pause-work", "stop work"	workflows/pause-work.md
"resume", "resume-work", "continue"	workflows/resume-work.md
"add todo", "add-todo", "capture"	workflows/add-todo.md
"check todos", "check-todos", "todos", "list todos"	workflows/check-todos.md
"add phase", "add-phase"	workflows/add-phase.md
"insert phase", "insert-phase", "urgent phase"	workflows/insert-phase.md
"remove phase", "remove-phase", "delete phase"	workflows/remove-phase.md
"new milestone", "new-milestone", "next milestone"	workflows/new-milestone.md
"complete milestone", "complete-milestone", "archive"	workflows/complete-milestone.md
"audit milestone", "audit-milestone", "audit"	workflows/audit-milestone.md
"settings", "config", "configure"	workflows/settings.md

</routing> <architecture> ## Workflow Files

Located in workflows/:

• new-project.md - Full project initialization workflow
• plan-phase.md - Phase planning with research and verification
• execute-phase.md - Wave-based execution orchestrator
• progress.md - Status check and intelligent routing to next action
• debug.md - Systematic debugging with persistent state
• quick.md - Ad-hoc tasks with GSD guarantees, skip optional agents
• discuss-phase.md - Gather context through adaptive questioning
• verify-work.md - Conversational UAT to validate built features
• map-codebase.md - Parallel codebase analysis for brownfield projects
• pause-work.md - Create handoff when pausing mid-phase
• resume-work.md - Resume with full context restoration
• add-todo.md - Capture ideas/tasks for later
• check-todos.md - List and work on pending todos
• add-phase.md - Add phase to end of milestone
• insert-phase.md - Insert urgent decimal phase
• remove-phase.md - Remove future phase and renumber
• new-milestone.md - Start new milestone cycle
• complete-milestone.md - Archive milestone and tag
• audit-milestone.md - Verify milestone completion
• settings.md - Configure workflow toggles

Agent Files

Located in agents/:

• gsd-project-researcher.md - Research domain ecosystem (stack, features, architecture, pitfalls)
• gsd-phase-researcher.md - Research how to implement a specific phase
• gsd-research-synthesizer.md - Synthesize parallel research into cohesive SUMMARY.md
• gsd-roadmapper.md - Create roadmap from requirements and research
• gsd-planner.md - Create detailed execution plans for a phase
• gsd-plan-checker.md - Verify plans will achieve phase goal before execution
• gsd-executor.md - Execute a single plan with task-by-task commits
• gsd-verifier.md - Verify phase goal achieved by checking must_haves against codebase
• gsd-debugger.md - Investigate bugs using scientific method with persistent state
• gsd-codebase-mapper.md - Analyze existing codebase for brownfield projects
• gsd-integration-checker.md - Verify cross-phase integration and E2E flows

Reference Files

Located in references/:

• questioning.md - Deep questioning techniques and context checklist
• ui-brand.md - UI/UX principles and brand guidelines

Templates

Located in templates/:

• project.md - PROJECT.md template
• requirements.md - REQUIREMENTS.md template
• research-project/ - Research output templates (STACK, FEATURES, ARCHITECTURE, PITFALLS, SUMMARY)

Workflow Pattern

GSD uses orchestrator + subagent pattern:

1. Orchestrator (workflow) - Stays in main context, spawns subagents, routes flow
2. Subagents (agents) - Fresh context, focused task, return structured result
3. Iteration - Verification loops (planner → checker → planner) until quality gates pass

This allows:

• Lean orchestrator context (~15%)
• Fresh context per subagent (100%)
• Parallel execution (4 researchers, multiple plans in wave)
• Verification before wasting execution time </architecture>

<success_criteria>

• User can initialize new projects via /gsd new-project
• Full workflow executes: questioning → research → requirements → roadmap
• Phase planning includes research and verification loop
• Phase execution uses wave-based parallelization
• Verification checks must_haves against actual code
• .planning/ directory structure created with all artifacts
• Clear next steps provided at each stage </success_criteria>