Herdr Council — Development skill for Claude Code
herdr plugin: ask every coding agent the same question, then let a judge weigh the answers blind.
How to install Herdr Council
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open zekierman/herdr-council and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Herdr Council does
herdr plugin: ask every coding agent the same question, then let a judge weigh the answers blind.
Alternatives in Development
- Rb Ask — Ask a question about the current project's codebase via the repobrain knowledge hub 1.3k ★
- Ag Ask — Ask a question about the current project's codebase via the antigravity knowledge hub 1.2k ★
- Herdr Reviewr — Review your coding agent's diff in a terminal pane and send line comments back to Claude Code, Codex, OpenCode 823 ★
README
Council

With sound:
https://github.com/user-attachments/assets/5e04b781-c49b-4102-8fe3-ec7930d38a13
A [herdr](https://github.com/herdrdev/herdr) plugin. Put one question to every coding agent in your workspace, let each answer on its own, then have a judge agent weigh the answers blind and write one verdict.
- Ask. Write the question once.
- Seats answer alone. Each selected agent (Claude Code, Codex, Antigravity, ...) gets the question in its own pane and writes its answer to its own folder. No seat sees another's answer.
- Peer review. Each agent ranks the other agents' answers, anonymised as A, B, C, and never its own. The rankings are averaged. (On by default for up to six seats; it doubles the time.)
- Blind verdict. A judge agent reads the anonymised answers and the peer ranking, with names hidden and order shuffled, and writes: consensus, conflicts, what was missed or risky, and a final answer. Then the names are revealed.
Different models notice different things. Where they agree you can trust the answer more; where they disagree there is usually a risk worth knowing about. Council works with whatever agents are running in the workspace, two or fifty.
See it work
A real run in herdr: three agents (Antigravity, Claude Code, Codex) review a small `retry.ts`, rank each other's answers, and Claude judges blind. Recorded with the mouse; the wait for answers is sped up 16×.
https://github.com/user-attachments/assets/123ce53b-2bc3-4830-9188-463ca6d69b86
Install
herdr plugin install zekierman/herdr-council
Requires herdr 0.9 or newer, on Windows, macOS or Linux. The install step downloads a prebuilt binary and checks its SHA-256; if that isn't possible it builds from source with Go.
Open it
The first time i
Related Skills
Kimi CLI Agent
Answers one council question as plain text, with no tool access.
Ask Council
Ask the council a question and get a synthesised answer from the appropriate panel of agents
Panel
Fan a question out to every available provider in parallel (read-only), then synthesize their answers against
Herdr Telegram Agents
Herdr plugin: a Telegram forum topic per live coding agent, status in the topic icon, prompts and answers both
Audio Compare
Advisory before/after audio judgment — measure → compare → judge two renders and weigh in on a DSP change with
Caffeine
Keep your Mac awake while you work, then let it sleep when your coding agent waits for you.
Related Agents
Council Opus
Council seat - intent and edge cases. Judges what the owner meant but did not write, evidencing every ask by r
Brain Eval Engineer
Evaluation engineer — question set tooling, layered metrics (harvest/graph/retrieval/answer), fixed-strategy a
Interview Panel
Generates and stress-tests interview answers from three independent perspectives (hiring manager, peer enginee