Skill

council-review-karpathy

Run a multi-persona LLM-council review using Andrej Karpathy's original llm-council Python implementation. Clones the repo into a temp scratch dir, configures it for the user's idea, runs it, and copies outputs back into the idea workspace. Use when the user explicitly wants the canonical implementation rather than the native subagent variant.

npx claudepluginhub danielrosehill/claude-code-plugins --plugin business-idea-eval

Tool Access

This skill uses the workspace's default tool permissions.

Preview

Wraps the upstream https://github.com/karpathy/llm-council implementation. We do not vendor the code — we clone it on demand into a scratch directory and run it from there.

SKILL.md

Similar Skills

skill-lookup

159.9k

Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.

prompts.chat

next-compile

139.2k

Checks Next.js compilation errors using a running Turbopack dev server after code edits. Fixes actionable issues before reporting complete. Replaces `next build`.

1 file

vercel-next-js-2

angular-new-app

100.1k

Creates new Angular apps using Angular CLI with flags for routing, SSR, SCSS, prefixes, and AI config. Follows best practices for modern TypeScript/Angular development. Use when starting Angular projects.

angular-angular-1

Stats

Stars0

Forks0

Last CommitApr 28, 2026

Actions

View Source View Plugin View on GitHub View README

Help us improve

Share bugs, ideas, or general feedback.

council-review-karpathy

Council Review (Karpathy)

Wraps the upstream https://github.com/karpathy/llm-council implementation. We do not vendor the code — we clone it on demand into a scratch directory and run it from there.

Steps

Verify prerequisites: python3, git, and an LLM API key the upstream tool supports (typically OPENAI_API_KEY or ANTHROPIC_API_KEY — check upstream README at clone time).

Choose a scratch directory:

SCRATCH="${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/business-idea-eval/scratch/llm-council"
mkdir -p "$(dirname "$SCRATCH")"

Clone (or pull) the upstream repo into $SCRATCH. Don't commit it anywhere; this is throwaway.

Read $SCRATCH/README.md (it may have changed since this skill was written) and follow its current setup steps. Do not assume specific commands — read what's there.

Prepare the prompt input from the idea workspace:

Concatenate idea.md + the formatted summary version (if present) + a short user-context paragraph
Save as the input file in whatever format the upstream tool expects

Configure the council personas to match the six in reference/council-deliberation.md. The upstream tool typically accepts custom persona prompts via a config file.

Run the council deliberation per upstream instructions.

Copy outputs into the idea workspace:

<workspace_path>/ideas/<slug>/council-karpathy/
  <whatever-files-the-tool-produces>

Optionally write a small index file listing the outputs and the run metadata (model used, timestamp, persona config).

Notes

The upstream tool may require API costs the user should be aware of — show the rough request count before running.

Do not commit the scratch clone to any repo. It exists only for the run.

If the upstream tool's interface has drifted, fall back to council-review-subagents and report the drift to the user.

Output schema differs from the subagent variant; downstream skills (synthesize-analysis, internal-pdf) handle both folder layouts.