mcpField
E
elon-musk

elon-musk

github.com/alpoxdevProductivity537 installsUpdated 2w ago
Repository

About elon-musk

elon-musk is a decision-making skill that uses first-principles analysis to challenge inherited assumptions and develop alternatives. It organizes research, assumptions, redesign options, and an action plan in a multi-file folder.

Key features

  1. 01Classifies assumptions with an A/B/C model
  2. 02Rebuilds options from fundamentals
  3. 03Stress-tests choices with inversion and pre-mortems
  4. 04Tracks phases in flow.json
  5. 05Saves results in separate files

Use cases

  1. 01Reconsidering crowded SaaS pricing
  2. 02Examining high infrastructure costs
  3. 03Finding alternatives when retention approaches fail

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/alpoxdev/hypercore --skill elon-musk

After installing, type/elon-muskin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

SKILL.ko.md
references/
flow-schema.md
frameworks.md
rules/
execution.md

Details

From the project README

@rules/execution.md @references/frameworks.md

Elon Musk

Rebuild decisions from first principles instead of inheriting analogy-driven defaults. Organize results as a multi-file folder for future reference.

<when_to_use>

Use this skill when:

  • the current options all look like slight variations of the same idea
  • cost, structure, or strategy feels constrained by habit
  • you need to separate real constraints from inherited assumptions

Do not use this skill when:

  • the job is mostly factual research with no redesign
  • the request is standard implementation planning
  • the user needs startup validation scoring rather than assumption teardown

Boundary:

  • if the main job is startup scoring and readiness assessment, use startup-validator
  • if the main job is broad idea generation, use genius-thinking

Examples:

/elon-musk SaaS pricing is crowded and undifferentiated
/elon-musk infrastructure cost is 40% of revenue
/elon-musk churn is high and retention playbooks are failing

</when_to_use>

<input_check>

If $ARGUMENTS is missing, ask:

Which problem should we deconstruct using first principles?

</input_check>

<owned_job>

For each run:

  1. Restate the problem and desired outcome.
  2. Gather facts only where the answer depends on current evidence.
  3. Classify assumptions using the A/B/C model from references/frameworks.md.
  4. Rebuild options from the surviving fundamentals.
  5. Stress-test the best path with inversion and pre-mortem.

</owned_job>

<document_shape>

Output Structure

.hypercore/elon-musk/[topic-slug]/
├── flow.json           # phase tracking
├── research.md         # domain research (conventions + facts + innovation cases)
├── assumptions.md      # A/B/C matrix with Socratic questioning
├── redesign.md         # rebuilt options from fundamentals + comparison table
└── execution.md        # inversion + pre-mortem + action plan
  • Use ASCII kebab-case for [topic-slug] (e.g., saas-infra-cost).
  • Each phase produces its own file for organized reference.
  • flow.json tracks progress through phases. See references/flow-schema.md for the schema.
  • If the folder exists from a prior run, read existing files before updating.

</document_shape>

<flow_tracking>

Flow Tracking

Write flow.json at the start and update as each phase completes.

Phase progression
PhaseOutput fileNext
researchresearch.md — industry conventions, actual facts, innovation casesdeconstruct
deconstructassumptions.md — A/B/C matrix with Socratic questioningredesign
redesignredesign.md — rebuilt options + current vs first-principles comparisonexecute
executeexecution.md — inversion, pre-mortem, action sequencedone
Resume support

If flow.json already exists, read it and continue from the last incomplete phase. Do not restart completed phases.

</flow_tracking>

Workflow

PhaseTaskOutput file
1Clarify the decision, gather domain researchresearch.md
2Identify constraints, conventions, unknowns (A/B/C matrix)assumptions.md
3Rebuild alternatives from fundamentalsredesign.md
4Stress-test the best path, recommend next actionsexecution.md

Research rule:

  • if the problem depends on current facts, gather and cite them
  • if the problem is conceptual, do not force web or team workflows

<output_contract>

Each output file should include:

  • research.md: industry conventions, actual data/benchmarks, innovation cases (with source URLs)
  • assumptions.md: A/B/C matrix (A=physical constraints, B=conventions to eliminate, C=needs verification), Socratic questioning applied
  • redesign.md: current vs first-principles comparison table, 3-5 alternative paths with feasibility/impact scores
  • execution.md: inversion failure scenarios (5-7), pre-mortem analysis, prioritized action sequence

</output_contract>

Before finishing, check:

  • conventions are separated from true constraints
  • unknowns that need evidence are labeled explicitly
  • the recommendation does not depend on obsolete tool or model names
  • inversion or pre-mortem was used to challenge the preferred path
  • all output files are saved under .hypercore/elon-musk/[topic-slug]/
  • flow.json status is set to completed

Comments

Markdown supported