gabelul

Bpx Council — AI skill for Claude Code

AI community

A portable multi-model council CLI — one advisor or a whole debating council, usable from any coding agent (Claude Code, Codex, Cursor, pi).

How to install Bpx Council

This entry records only its repository, not the path inside it, so there is no exact command to give. Open gabelul/bpx-council and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Bpx Council does

A portable multi-model council CLI — one advisor or a whole debating council, usable from any coding agent (Claude Code, Codex, Cursor, pi).

Alternatives in AI

Codeburn — Free, local tool to track AI coding token usage and cost across 37 tools and agents (Claude Code, Cursor, Code 9.7k ★
Skills Manage — Desktop app to manage AI coding agent skills across Claude Code, Cursor, Gemini CLI, Codex, and 20+ platforms 2.2k ★
Token Monitor — Local-first desktop widget for tracking token usage, costs, and limits across 32+ AI coding tools—including Cl 1.8k ★

README

bpx-council

Multi-model second opinions for coding agents.

Your agent runs on a fast, cheap model most of the time — right, most of the time. The exceptions are the handful of calls that decide how the thing turns out: the architecture, the "should I even build this," the bug it's been circling for an hour. `bpx-council` puts a stronger model on those. Or three different ones, arguing.

What it looks like

Real output, unedited. One question, three models, each with a stance:

bpx-council --mode council --backends codex,claude,opencode \
  "Our test suite takes 40 minutes and it's blocking deploys. Parallelize
   across more CI runners, or delete slow integration tests we suspect
   are redundant?"

── council: architect→codex, critic→claude, simplifier→opencode

architect [for] · codex

Parallelize the suite across more CI runners now. Do not delete suspected-redundant integration tests as the primary fix. A 40-minute deploy gate is an infrastructure throughput problem first. […] Split tests into deterministic shards using historical timing, targeting 8–12 minutes per shard. Balance by measured duration, not test count.

critic [against] · claude

**Parallelization is a treadmill with a floor you haven't measured.** Amdahl's law eats this alive. If one integration test file takes 11 minutes, sharding across 20 runners gets you an 11-minute suite, not a 2-minute one. […] "We suspect are redundant" is not evidence, and the failure mode is invisible. Delete a test, ship for six months, eat a production incident — nobody ever traces that back to the deletion.

**Both proposals are answers to "make the number smaller." Neither is an answer to "stop the number from gating deploys."**

Verdict

Don't delete suspected-redundant integration tests, and don't make runner sharding the first move. […] My call: **take the critic's pipeline change first, then selectively optimize and shard.**

Two models