gsd-orchestrator
GSD Orchestrator
GSD v1.34.2 — tools at ~/.claude/get-shit-done/
GSD bin: node "$HOME/.claude/get-shit-done/bin/gsd-tools.cjs"
Core Principle
Two modes in one skill: consultant (advise the right path) and orchestrator (drive the workflow). Always orient first, then advise, then act. Never lose the thread — show status on every response. Ask when unclear rather than guess.
Step 0: Orient Before Everything
Run this before any user interaction:
node "$HOME/.claude/get-shit-done/bin/gsd-tools.cjs" init progress 2>/dev/null
Then show:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
GSD | Status: [new / phase N of M / milestone complete]
| Project: [name or "none"]
| Mode: [interactive / yolo / unknown]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Step 1: Assess Intent & Advise
Before running anything, assess what the user actually needs. If the request is clear, proceed. If not, ask — max 2 targeted questions.
Clarity check
| Signal | Response |
|---|---|
| User describes a vague task | Ask: "Is this a new project, a feature on an existing one, or a quick fix?" |
| Task size is unclear | Ask: "Roughly how big is this — a few files, a feature, or a full project?" |
| User doesn't know which GSD command | Use the Decision Framework below to recommend one |
| User asks "do I need GSD for this?" | Assess honestly — sometimes /gsd-fast or no GSD at all is right |
Decision Framework
| Task size | Right tool |
|---|---|
| < 3 files, obvious | /gsd-fast "description" — inline, zero overhead |
| Small, wants tracking | /gsd-quick — plan + execute, no subagents |
| Small + needs QA | /gsd-quick --validate — adds checking + verification |
| Non-trivial feature | Full flow: discuss → plan-phase → execute-phase |
| New project from scratch | /gsd-new-project |
| New milestone on existing | /gsd-new-milestone "name" |
| Bug | /gsd-debug "description" — persists across /clear |
| Urgent mid-milestone work | /gsd-insert-phase N "description" |
| Want to see plan before coding | /gsd-list-phase-assumptions N |
| Full autonomous run | /gsd-autonomous |
When to use discuss-phase?
Before /gsd-plan-phase N when the user has opinions about the approach, UX choices, or architecture. Skip for mechanical phases.
Model profile advice
| Profile | Use when |
|---|---|
quality |
Complex architecture, greenfield, high-stakes |
balanced |
Standard feature work (default) |
budget |
Boilerplate, docs, low-stakes |
inherit |
Match current session model |
Step 2: Execute the Workflow
Once the path is clear, drive it. Show the status header before every response.
Route A: New Project
/gsd-new-project(or--auto @doc.mdif user has a PRD)- Confirm REQUIREMENTS.md + ROADMAP.md created
- Recommend
/gsd-plan-phase 1→ Route B
Route B: Plan a Phase
- Offer
/gsd-discuss-phase Nif user has opinions on the approach /gsd-plan-phase N- Show PLAN.md summary — ask: "Ready to execute, or adjust first?"
- → Route C
Route C: Execute a Phase
/gsd-execute-phase N- Show SUMMARY.md results
/gsd-verify-work N— surface failures- Failures →
/gsd-debugeach blocker → re-verify - Green → ask: "Ship (
/gsd-ship N) or next phase?" - → Route B or Route D
Route D: Milestone Completion
/gsd-audit-milestone→ show gaps- Gaps →
/gsd-plan-milestone-gaps→ Route B for fix phases - Clean →
/gsd-complete-milestone <version> - Ask: "New milestone or done?"
Status Header (every response)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Phase: N/M — [name] Mode: [interactive/yolo]
Done: ✓ p1 ✓ p2 … Next: p(N+1), p(N+2)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Subagent Dispatch Template
Execute this GSD step in: [working directory]
Step: [gsd command + args]
State context: [paste relevant STATE.md excerpt]
Expected output: [PLAN.md / SUMMARY.md / VERIFICATION.md]
Output when done:
RESULT: <one sentence>
ARTIFACTS: <files created/modified>
BLOCKERS: <decisions needing human input>
Anti-patterns to Flag
/gsd-execute-phasewithout a PLAN.md → pre-flight gate will fail. Plan first./gsd-fastfor multi-file refactors → redirect to/gsd-quick --validate.- New milestone without audit → suggest
/gsd-audit-milestonefirst. - Skipping
/clearbetween plan and execute on large phases → context bloat degrades agents.
Focus Recovery
If the conversation drifts:
"Noted — capturing as
/gsd-add-todo. We're on phase N. Continue, or handle this first?"
Never skip the status header. Never skip verification. Never advance without user confirmation.
More from tebjan/agent-skills
caveman-smart
Intelligent token-efficiency mode. Compresses routine chat and thinking but stays in full prose for reasoning, architecture, and safety-critical topics. Applies per response, not per session. Use when the user says 'caveman', 'terse', 'kurz', 'kurz&knapp', 'kurz bitte', 'weniger text', 'less tokens', 'smart mode', or invokes /caveman-smart. Stays active until the user says 'normal mode', 'verbose', or 'ausführlich'. Built for HPC, real-time, graphics, embedded, and kernel work where technical precision matters more than word count but explanation-of-why still demands full prose.
2fix-jsonl-surrogates
Diagnose and repair 'invalid high surrogate' API errors in Claude Code chat sessions. Use this skill when the user encounters API Error 400 with 'invalid high surrogate in string', 'not valid JSON', surrogate-related errors, or wants to fix/scan/repair a broken Claude Code JSONL chat file. Also use when a session refuses to resume due to JSON encoding errors, or when the user mentions request IDs (req_...) alongside JSON parse failures.
2swarm-advisor
Advises on multi-agent swarm design using Claude Code's TeammateTool and Task system. Use when the user wants to parallelize work across agents, asks 'should I use a swarm for this?', wants help designing team topology, needs to break a task into parallel workstreams, or is getting started with swarm-orchestration. Assesses whether a swarm is worth the overhead, recommends team size and structure, helps decompose tasks with correct dependencies, and tracks swarm health. Complements the swarm-orchestration skill by adding the 'think before you spawn' layer.
1install-github-plugin
Installs Claude Code plugins from GitHub repos that lack a marketplace manifest. Use this skill whenever a `claude plugin install` fails due to SSH key errors, missing `.claude-plugin/marketplace.json`, or a GitHub repo that wasn't designed as a Claude plugin marketplace. Also use when the user says \"install plugin from GitHub\", \"add skills from a repo\", or \"the plugin install failed\". Handles the full workflow: cloning via gh CLI, inspecting repo structure, generating the marketplace manifest, and running the install.
1addy-orchestrator
Orchestrates end-to-end software development using the addyosmani/agent-skills framework. Guides the user through define → plan → build → verify → review → ship phases, spawns subagents for each step, tracks state persistently, and never loses focus on workflow completion. Use when the user says \"let's build X\", \"help me implement X\", \"walk me through X\", or wants structured multi-phase dev guidance. Also triggers when a task is clearly non-trivial and would benefit from phased execution.
1ralph-orchestrator
Orchestrates and advises on Ralph Loop automated development sessions. Use when the user wants to run ralph-loop, asks how to set up a Ralph task, wants iterative agent-driven development, says 'run this in a loop until it works', or has a task with testable acceptance criteria they want to automate. Helps write effective loop prompts, sets safe iteration limits, monitors loop progress, detects stuck loops, and advises on completion criteria. Keeps context short — the heavy work happens inside the loop.
1