Transform a user's builder vision into a complete, executable GSD mission package. Use this skill whenever a user has described what they want to BUILD (a product, tool, feature, system, educational pack, or ecosystem component) and wants it structured for GSD execution. Triggers include: 'structure this for GSD', 'make this into a milestone', 'turn this vision into a mission', 'package this up for Claude Code', 'create the mission files', 'I want to hand this to GSD', 'make me a mission package', or any request to decompose a described system into wave-based executable tasks. Also trigger when the user has a vision doc already written and needs mission decomposition. Prefer this skill over generic document creation whenever the GSD ecosystem is mentioned alongside building something.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Transform a user's builder vision into a complete, executable GSD mission package. Use this skill whenever a user has described what they want to BUILD (a product, tool, feature, system, educational pack, or ecosystem component) and wants it structured for GSD execution. Triggers include: 'structure this for GSD', 'make this into a milestone', 'turn this vision into a mission', 'package this up for Claude Code', 'create the mission files', 'I want to hand this to GSD', 'make me a mission package', or any request to decompose a described system into wave-based executable tasks. Also trigger when the user has a vision doc already written and needs mission decomposition. Prefer this skill over generic document creation whenever the GSD ecosystem is mentioned alongside building something.
version
1.0.0
format
"2025-10-02T00:00:00.000Z"
triggers
["a user has described what they want to BUILD (a product, tool, feature, system, educational pack, or ecosystem component) and wants it structured for GSD execut","'structure this for GSD', 'make this into a milestone', 'turn this vision into a mission', 'package this up for Claude Code', 'create the mission files', 'I wan","the user has a vision doc already written and needs mission decomposition"]
updated
"2026-04-25T00:00:00.000Z"
status
ACTIVE
Vision โ Mission Skill
Transform a user's builder vision โ described in conversation โ into a complete, GSD-ready mission package: a structured folder of markdown files ready for handoff to Claude Code and the GSD orchestrator.
This skill produces builders' output, not researchers' output. It generates the executable files: milestone spec, component specs, wave plan, test plan, and README. Compare with research-mission-generator, which produces a LaTeX research document. Use this skill when the goal is to build something.
Pipeline Speed Detection
Detect which pipeline the conversation warrants โ do not ask the user unless genuinely ambiguous.
Speed
When
Stages Produced
Full (Vision โ Research โ Mission)
Domain requires professional accuracy, safety, or cultural sensitivity (electronics, nutrition, cultural heritage, medical)
All three stages
Fast (Vision + Mission, skip research)
Domain within Claude's training, pure software/tooling, internal GSD components
Vision doc + mission files
Mission-only
User already has a vision doc, needs decomposition
Mission files only
Step 1: Harvest the Conversation
Scan the current conversation (and past conversations if referenced) for:
User stream-of-consciousness while Claude listened
Highest priority โ preserve verbatim insights
Critical: Preserve the user's own language. Their metaphors, framings, and analogies belong in the vision narrative. Don't replace user thinking with generic content.
Step 2: Identify the Archetype
Read references/vision-archetypes.md to classify the vision. Each archetype has characteristic components, common failure modes, and chipset patterns. Classification shapes the output structure.
Quick reference:
Archetype
Key Signals
Mission Shape
Educational Pack
Curriculum, modules, Try Sessions, cultural content
Skip if the user already has a vision doc. Otherwise, write 01-vision-doc.md from harvested content.
See references/mission-templates.md โ Vision Document Template for the complete structure.
Vision doc quality gates (run before proceeding):
Narrative makes someone want to build this (not a spec โ a destination)
Problem statement has 3โ5 concrete, recognizable problems
Core concept described in one interaction arc sentence
ASCII architecture diagram present
Chipset YAML configures the actual agents needed
Success criteria are testable by an agent (8โ12 criteria)
No function signatures in the vision (those belong in component specs)
Through-line connects to a GSD ecosystem principle
Step 4: Research Reference (Full Pipeline Only)
For Full pipeline speed only. Otherwise skip.
Conduct targeted web research using these source quality rules (ABSOLUTE):
Government agencies (USGS, NIH, NASA, EPA, NIST, etc.)
Peer-reviewed journals and university research
Professional organizations and official standards bodies
NEVER entertainment media, blogs, or unsourced claims
Structure as 02-research-reference.md. See references/mission-templates.md โ Research Reference Template.
Step 5: Decompose into Components
Identify 4โ12 components based on complexity. Each component maps to one Claude Code agent invocation. Every component spec must pass the self-containment test: delete every other file, hand an agent this one spec โ can they build it? Yes = done. No = copy more context in.
Component sizing guide:
Complexity
Components
Examples
Simple feature
4โ5
Dashboard widget, API endpoint
Medium system
6โ8
Plugin architecture, persona system
Complex pack
8โ12
Educational pack, full skill suite
Mega-milestone
12โ20
Electronics course, foundational knowledge suite
Component decomposition principles:
One component = one agent invocation = one verifiable artifact
Components have clear inputs (what they receive) and outputs (what they produce)
Shared types and schemas are always Wave 0, separate from implementation components
Integration and verification are always final wave, never parallelized with build work
Step 6: Assign Models
Apply the 30/60/10 principle. If >40% requires Opus, the vision needs further decomposition.
Complexity Signal
Model
Examples
Judgment, creativity, architecture
Opus
Personality design, safety warden logic, activation engines, cultural sensitivity, factory/meta-systems
Structural implementation
Sonnet
Schema definitions, pipelines, registries, test suites, documentation, content generation from patterns
Scaffold and boilerplate
Haiku
Directory structures, config files, type stubs, templates, simple transformations
Step 7: Plan Waves
See references/mission-templates.md โ Wave Execution Plan Template for the output format.
Wave planning principles:
Wave 0 always exists โ shared types, schemas, interfaces; must complete within 5-min cache TTL
Push to parallel โ any components without shared state should run simultaneously
Safety is never parallel โ Safety Warden is always on the critical path, final wave
Integration after parallel โ Wave N-1 builds; Wave N integrates and verifies
cd /home/claude && zip -r [milestone-name]-mission.zip [milestone-name]-mission/
cp [milestone-name]-mission.zip /mnt/user-data/outputs/
Step 10: Deliver
Present the zip file and a brief summary. Keep it short โ the user can read the documents. Mention:
Total component count and wave structure
Activation profile (Patrol/Squadron/Fleet)
Opus/Sonnet/Haiku split
How to hand it to Claude Code (claude --new-task or GSD new-project)
Ecosystem Patterns (Mandatory โ Apply to All Output)
Mission Control Pattern
"We are the architects. Claude Code is the astronaut. Each mission doc is a complete flight plan." Every component spec self-contained. No spec may say "see the other specs."
Safety Warden Pattern
Three modes: annotate (flag for awareness), gate (confirm before proceeding), redirect (block + suggest alternative). Safety monotonicity: children never relax parent safety boundaries.
Progressive Disclosure Pattern
Three reading speeds in all output: glance (title + summary line), scan (headers + tables), read (full body).
Humane Flow Pattern
Systems support, never shame. No guilt messaging. The system says "Welcome back."
Amiga Principle
Remarkable results through architectural intelligence, not raw power. Specialized execution paths. Every architectural decision should ask: "Is there a clever structural solution that makes the brute-force approach unnecessary?"
Common Mistakes to Avoid
Vision docs that are specs. Vision = destination. No function signatures.
Missions requiring other missions. Each component is self-contained. Period.
Everything assigned to Opus. Most work is structural. Save Opus for judgment.
Sequential waves that could be parallel. No shared state โ parallelize.
Missing the cache window. Wave 0 must complete in <5 min.
Test plans mirroring criteria 1:1. Tests โฅ criteria ร 2.
Skipping the through-line. Keeps the milestone aligned with GSD values.
Generic vision narratives. The user's own metaphors must appear in the vision.
Reference Files
references/vision-archetypes.md โ Detailed archetype patterns, characteristic components, chipset YAML patterns, common failure modes. Read for complex systems before writing the vision doc.
references/mission-templates.md โ Complete templates for every output file (Vision Doc, Research Reference, Milestone Spec, Component Spec, Wave Execution Plan, Test Plan, README). Read before generating any mission file.
