Skip to content
Back to search

autodecision

A decision operating system for high-stakes choices — business, strategy, career. Simulates disagreement, stress-tests assumptions, and converges on what actually holds up. Claude Code skill inspired by Karpathy's autoresearch + LLM council.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/harshilmathur--autodecision/badge.svg)](/listings/harshilmathur--autodecision)

HTML

<a href="/listings/harshilmathur--autodecision"><img src="/api/listings/harshilmathur--autodecision/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Works with [Claude Code](https://claude.ai/code) and [Claude Cowork](https://claude.com/product/cowork).

In Claude Code

/plugin marketplace add harshilmathur/autodecision
/plugin install autodecision@autodecision

Commands land under `/autodecision:`. The main loop is `/autodecision:autodecision`. Subcommands share the same prefix.

In Claude Cowork

1. **Customize → Create plugin → Add marketplace**, paste `https://github.com/harshilmathur/autodecision` 2. **Customize → Add plugin → Personal → autodecision → Install**

<details> <summary>Other install paths (legacy skill, offline zip)</summary>

**Legacy skill install (Claude Code).** Copies skill files directly into `~/.claude/`:

git clone https://github.com/harshilmathur/autodecision.git
cd autodecision
./install.sh

Bare `/autodecision "..."` works in this mode. If both paths are installed, the plugin wins.

**From release zip (Cowork, offline).** Download `autodecision-<version>.zip` from the [latest release](https://github.com/harshilmathur/autodecision/releases/latest), then: **Customize → Create plugin → Upload plugin**, select the zip.

</details>

Then run

/autodecision:autodecision "Should we cut pricing by 20%?"
/autodecision:quick "Should we launch in Southeast Asia?"
/autodecision:challenge "We're dropping UPI fees to zero next month"

The system scopes, grounds, simulates, critiques, stress-tests, and produces a Decision Brief.

---

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 23 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. {
  2. "name": "autodecision",
  3. "version": "0.4.0",
  4. "owner": {
  5. "name": "Harshil Mathur",
  6. "url": "https://github.com/harshilmathur"
  7. },
  8. "plugins": [
  9. {
  10. "name": "autodecision",
  11. "source": "./claude-plugin",
  12. "description": "Iterative decision simulation with a 5-persona council, adversarial red-teaming, and mechanical convergence. Applies Karpathy's autoresearch + LLM Council patterns to high-stakes business and strategy decisions.",
  13. "version": "0.4.0",
  14. "category": "productivity",
  15. "keywords": [
  16. "decision-making",
  17. "strategy",
  18. "simulation",
  19. "adversarial"
  20. ]
  21. }
  22. ]
  23. }
README.md · README · 230 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <p align="center">
  2. <h1 align="center">autodecision</h1>
  3. <p align="center">
  4. <b>A decision operating system for high-stakes choices.</b> Business, strategy, career.<br />
  5. Simulates disagreement, stress-tests assumptions, and converges on what actually holds up.
  6. <br /><br />
  7. <sub>Applies <a href="https://github.com/karpathy/autoresearch">Karpathy's autoresearch</a> + <a href="https://github.com/karpathy/llm-council">LLM Council</a> patterns to decisions.</sub>
  8. <br /><br />
  9. <a href="#quick-start">Quick Start</a> · <a href="#how-it-works">How It Works</a> · <a href="#examples">Examples</a> · <a href="#commands">Commands</a> · <a href="TODOS.md">Roadmap</a>
  10. </p>
  11. </p>
  12. <p align="center">
  13. <img src="docs/effects-viz-preview.gif" alt="Orbital effects map: hypotheses on the inner ring, 1st- and 2nd-order effects fanning outward, worst-case diamonds and black-swan comets on the outer rings" width="900">
  14. <br />
  15. <sub>Static preview of the full possibility map for one decision: 5 hypotheses, ~110 effects, worst cases, black swans. Interactive viewer in development.</sub>
  16. </p>
  17. ---
  18. ## Why decisions fail
  19. Most bad decisions don't look bad upfront. They fail later, in second-order effects, edge cases, and under stress.
  20. People routinely:
  21. - Think in **first-order effects** only ("cut prices, sales go up")
  22. - Ignore **second-order effects** ("…and then competitors match, margins compress, existing customers renegotiate, NRR drops")
  23. - Underestimate **worst-case scenarios** ("what if we're wrong by 2x?")
  24. - Miss **black swan events** ("what if visa risk and market risk are correlated?")
  25. By the time these failure modes appear, the decision is already in motion and hard to reverse.
  26. Autodecision exists because these failure modes are predictable, if you force yourself to look for them.
  27. ---
  28. ## How it works
  29. ```
  30. OUTER (runs once):
  31. Phase 0 SCOPE Decompose decision → sub-questions
  32. Phase 1 GROUND Web search for real data and precedents
  33. Phase 1.5 ELICIT Review assumptions, personas, data with user
  34. INNER (iterates until convergence, default 2x):
  35. ┌──────────────────────────────────────────────────────┐
  36. │ Phase 2 HYPOTHESIZE Generate competing paths │
  37. │ Phase 3 SIMULATE 5 parallel persona agents │
  38. │ Phase 4 CRITIQUE Anonymized peer review │
  39. │ Phase 5 ADVERSARY Red-team + black swan tests │
  40. │ Phase 6 SENSITIVITY Find decision boundaries │
  41. │ Phase 7 CONVERGE Judge measures stability │
  42. └──────────────────────────────────────────────────────┘
  43. OUTER (runs once):
  44. Phase 8 DECIDE Produce Decision Brief
  45. ```
  46. The 5 personas run as parallel subagents, each with its own context window, genuinely unable to see the others:
  47. | Persona | Sees | Blind Spot |
  48. |---------|------|------------|
  49. | Growth Optimist | Upside, creative alternatives | Execution risk |
  50. | Risk Pessimist | Downside, failure modes | Opportunity cost of inaction |
  51. | Competitor Strategist | Market dynamics, game theory | Overestimates rationality |
  52. | Regulator | Legal, compliance, constraints | Overweights unlikely regulation |
  53. | Customer Advocate | User value, adoption, retention | Ignores unit economics |
  54. The output is a 16-section Decision Brief, optionally renderable as a McKinsey-styled 16-slide PPTX deck via `/autodecision:deck`. Every probability comes with a [min, max] range reflecting council disagreement. Every fragile insight comes with the exact threshold where it flips. Every dollar figure carries a source tag. See a full brief and its deck: [Law Firm AI Replacement (brief)](examples/law-firm-ai-replacement.md) · [Deck (PDF)](examples/decks/04-law-firm-1Y-replacement.pdf).
  55. If iterations hit the cap without converging, the orchestrator pauses, shows the Judge scores, and asks if you want another pass (one-at-a-time, cap 5 total). Never silently exits with "NOT REACHED."
  56. ---
  57. ## Best for
  58. - **Strategic business decisions**: pricing, expansion, M&A, capital allocation, build-vs-buy, senior hires
  59. - **Career decisions with asymmetric outcomes**: role moves, relocation, founder vs operator, fundraising terms
  60. - **Situations with second-order effects**: where the obvious answer isn't the robust answer
  61. - **Hard-to-reverse decisions**: where you only get one shot and want to stress-test it first
  62. Skip it for simple factual questions (one LLM call is fine), low-stakes everyday choices (overhead isn't worth it), and decisions you already have high conviction on (this will just slow you down).
  63. ---
  64. ## Quick start
  65. Works with [Claude Code](https://claude.ai/code) and [Claude Cowork](https://claude.com/product/cowork).
  66. ### In Claude Code
  67. ```
  68. /plugin marketplace add harshilmathur/autodecision
  69. /plugin install autodecision@autodecision
  70. ```
  71. Commands land under `/autodecision:`. The main loop is `/autodecision:autodecision`. Subcommands share the same prefix.
  72. ### In Claude Cowork
  73. 1. **Customize → Create plugin → Add marketplace**, paste `https://github.com/harshilmathur/autodecision`
  74. 2. **Customize → Add plugin → Personal → autodecision → Install**
  75. <details>
  76. <summary>Other install paths (legacy skill, offline zip)</summary>
  77. **Legacy skill install (Claude Code).** Copies skill files directly into `~/.claude/`:
  78. ```bash
  79. git clone https://github.com/harshilmathur/autodecision.git
  80. cd autodecision
  81. ./install.sh
  82. ```
  83. Bare `/autodecision "..."` works in this mode. If both paths are installed, the plugin wins.
  84. **From release zip (Cowork, offline).** Download `autodecision-<version>.zip` from the [latest release](https://github.com/harshilmathur/autodecision/releases/latest), then: **Customize → Create plugin → Upload plugin**, select the zip.
  85. </details>
  86. ### Then run
  87. ```
  88. /autodecision:autodecision "Should we cut pricing by 20%?"
  89. /autodecision:quick "Should we launch in Southeast Asia?"
  90. /autodecision:challenge "We're dropping UPI fees to zero next month"
  91. ```
  92. The system scopes, grounds, simulates, critiques, stress-tests, and produces a Decision Brief.
  93. ---
  94. ## Examples
  95. ### Full loop (5 personas, 2 iterations, adversarial stress-testing)
  96. **[Law Firm AI Replacement](examples/law-firm-ai-replacement.md)** · [Deck](examples/decks/04-law-firm-1Y-replacement.pdf): Should a mid-sized law firm replace all first-year associates with Claude + senior review?
  97. - Full replacement was **unanimously ranked last**: pipeline rupture in year 3, malpractice risk, and ranking damage all compound
  98. - The winning recommendation **didn't exist in any single persona's output**. It was synthesized from three independent alternatives (pessimist + customer advocate + regulator)
  99. - **Regulator surfaced Utah vs Arizona sandbox distinction**, WARN Act compliance, and MRPC 5.3 supervision analysis
  100. - Final: a **wrapped 18-month pilot** cutting the 1Y class 50-60% with binding kill criteria, anchor-client co-design, and regulatory sandbox cover
  101. **[Buy vs Rent vs Relocate](examples/buy-vs-rent-vs-relocate.md)** · [Deck](examples/decks/05-buy-vs-rent-vs-sf.pdf): A dual-tech-income couple evaluates buying a house in Bangalore (10 Cr), renting + investing the delta, or relocating to San Francisco.
  102. - Council surfaced a **creative alternative** (buy small + invest the rest) that outperformed all three original options on stress resilience
  103. - **SF savings corrected** from $150-250K to $80-120K after modeling actual CA taxes, childcare, and living costs
  104. - **Adversarial red team** exposed a correlated risk: the scenario where you lose H1B sponsorship IS the scenario where your Indian portfolio is also down 30%
  105. - Final: a **staged real-options approach** with buy-small as the default endpoint
  106. **[10% vs 30% Price Cut Comparison](examples/comparison-10pct-vs-30pct.md)** (compare mode) · Decks: [10%](examples/decks/06-b2b-saas-pricing-cut-10pct-quick.pdf) · [30%](examples/decks/07-b2b-saas-pricing-cut-30pct-quick.pdf): Both reach "don't cut" but for **fundamentally different reasons**: 10% fails by being too small, 30% by being too large. Both converge on the same alternative.
  107. ### → See the full gallery: [EXAMPLES.md](EXAMPLES.md)
  108. Prompt patterns, decision types, and additional briefs across pricing, hiring, fundraising, build-vs-buy, and market expansion.
  109. ### Try these
  110. ```
  111. /autodecision "Should Adobe go all-in on agentic Creative Cloud and deprecate Photoshop's UI-first model within 3 years?"
  112. /autodecision "Should Uber build their own autonomous vehicles instead of partnering with Waymo/Cruise?"
  113. /autodecision "Should Netflix launch a free ad-supported tier in India, Brazil, and Indonesia?"
  114. /autodecision:quick "Should we cut pricing by 20%?"
  115. /autodecision "Should we cut pricing by 20%?" # compare quick vs full on the same question
  116. ```
  117. ---
  118. ## Commands
  119. | Command | What | Time |
  120. |---------|------|------|
  121. | `/autodecision` | Full loop, 5 personas, 2 iterations, convergence | ~15 min |
  122. | `/autodecision:quick` | Single-pass, no council | ~2 min |
  123. | `/autodecision:challenge` | Adversary-only stress test of a proposed action | ~5 min |
  124. | `/autodecision:compare` | Side-by-side comparison of two decisions | ~5 min |
  125. | `/autodecision:revise` | What-if on an existing run (changed assumptions/data) | ~8 min |
  126. | `/autodecision:publish` | Ship a brief as PDF → Notion, email, gist, Slack, Drive | ~1 min |
  127. | `/autodecision:deck` | Render a brief as a McKinsey-styled 16-slide PPTX | ~1 min |
  128. Pre-built templates: `--template pricing | expansion | build-vs-buy | hiring`.
  129. <details>
  130. <summary>More commands and flags</summary>
  131. **Other commands:**
  132. | Command | What | Time |
  133. |---------|------|------|
  134. | `/autodecision:summarize` | One-page shareable summary | ~1 min |
  135. | `/autodecision:plan` | Interactive scope wizard | ~2 min |
  136. | `/autodecision:review` | Past decisions + outcome tracking | ~1 min |
  137. | `/autodecision:export` | Portable archive of all decisions | ~1 min |
  138. **Iteration depth:**
  139. ```
  140. /autodecision --iterations 1 "decision" # Medium: council, 1 pass
  141. /autodecision --iterations 3 "decision" # Deep: up to 3 iterations
  142. ```
  143. **Skip the user review step** (Phase 1.5 ELICIT):
  144. ```
  145. /autodecision --skip-elicit "decision"
  146. ```
  147. **Attach context documents** (Claude Code only, supports `.md`, `.txt`, `.pdf`, `.csv`, `.json`, images):
  148. ```
  149. /autodecision "Should we take the Series A?" --context term-sheet.pdf
  150. /autodecision "Should we acquire Acme?" --context financials.csv analysis.md
  151. ```
  152. The engine extracts key data points, tags them `[D#]`, and threads them through the full pipeline alongside web-searched ground data. In Cowork, paste document content during ELICIT instead.
  153. </details>
  154. ---
  155. ## Data storage
  156. All decisions live in `~/.autodecision/runs/{slug}/` (user-level, never in your repo). A cross-decision `journal.jsonl` tracks outcomes; `assumptions.jsonl` tracks which assumptions held or broke across decisions, it compounds over time.
  157. ---
  158. ## License
  159. [MIT](LICENSE) · Roadmap in [TODOS.md](TODOS.md) · Contributing guide in [CONTRIBUTING.md](CONTRIBUTING.md)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.