[External Plugin]: agentcouncil
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 2/5
- 预计耗时
- 1-3 小时
- 新手友好度
- 55/100
- Issue 类型
- 功能
- 描述清晰度
- 基本清楚
- 活跃度
- 活跃
- 技术栈
- javascript
- 领域
- content
调研方向
Review the submission against the repository’s plugin submission requirements and inspect the referenced AgentCouncil release at commit abad48ddd3117056bbda12f2f7e87142d8c43369. Start with the plugin manifest, skill, and custom agent named in the issue; check that the listing details match the submitted repository and ref. Done when the submission is reviewed and any required corrections or decision are recorded.
由索引模型根据 Issue 内容生成。
描述
Plugin name
agentcouncil
Short description
Multi-model council for Copilot CLI: three subagents on different model families (Anthropic, OpenAI, Google) answer independently, then build on each other or attack the leading answer before an orchestrator writes the final result from anonymized inputs.
GitHub repository
Sentry01/AgentCouncil
Plugin path inside the repository
No response
Ref to review
v2.0.0
Commit SHA to review
abad48ddd3117056bbda12f2f7e87142d8c43369
Version
2.0.0
License identifier
MIT
Author name
Shlomi Shaki
Author URL
Homepage URL
No response
Keywords
multi-agent, multi-model, council, debate, brainstorm, stress-test, adversarial, collaborative, orchestration, second-opinion
Additional notes for reviewers
Not a duplicate of agent-council. The existing agent-council listing (Avyayalaya/agent-council) is a different project: a 5-agent quality gate that returns SHIP / REVISE / HOLD on text artifacts. AgentCouncil has a separate codebase, purpose and protocol. It answers open questions (decisions, designs, reviews, research) by seating one model from each of three families and making them interact once. The agentcouncil name keeps it distinct from the existing listing's name.
What it adds beyond a single model:
- Copilot CLI's
tasktool can run each subagent on a different model. AgentCouncil uses this to seat Anthropic, OpenAI and Google models side by side, so mistakes specific to one model family get caught. - The orchestrator only sees anonymized, reordered responses, which removes the bias toward the seat that shares its family.
- Two modes: collaborative (draft, improve, synthesize) and adversarial (draft, triage, attack, verdict). A light path covers simple questions in 4 calls.
- If a model is unavailable, the seat falls back to the next one on its list without collapsing two seats into one family. Every answer ends with a footer listing the models that actually ran.
- Every subagent prompt tells it to work read-only.
Contents:
- An Agent Plugins 1.0
plugin.json. - One skill:
skills/agentcouncil/SKILL.md, a short router, with phase prompts inreferences/. - One thin custom agent in
com.github.copilot/agents/, which loads the skill. There are no MCP servers, hooks or scripts that run at install or runtime.
Checks run before submitting:
@microsoft/vally0.18.0 lint: passed, covering spec-compliance and valid-refs.- In a clean
COPILOT_HOME, these work against the GitHub repo:copilot plugin marketplace add Sentry01/AgentCouncil,copilot plugin install agentcouncil@agentcouncilandcopilot plugin update.--agent agentcouncil:agentcouncilresponds. - Routing evals (22 cases, in the repo under
evals/): 44/44 over 2 runs on claude-sonnet-5.5, and 22/22 each on gpt-5.5 and gemini-3.8-flash. - End-to-end light and full adversarial councils ran on the default roster.
Releases are published from CHANGELOG.md. The repo is also its own marketplace (.github/plugin/marketplace.json).
Submission checklist
- The plugin lives in a public GitHub repository.
- The ref and/or sha I provided is immutable (release tag and/or full 40-character commit SHA), not a branch.
- This submission follows this repository's contribution, security, and responsible AI policies.
- This plugin is not already listed in the Awesome Copilot marketplace.
- 主要语言
- JavaScript
- 星标
- 40k
- 派生
- 5.1k
- 平均合并
- 2 天 10 小时
- 30 天内合并 PR
- 109
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
github/awesome-copilot 的其他 Issue
-
external-plugin needs-review:HIGH ready-for-review
难度 2/5 1-3 小时 新手友好度 70/100
github/awesome-copilot#5162 · 3 条评论 ·
维护者通常 1 天内回复
-
external-plugin external-plugin-canvas needs-review:MEDIUM ready-for-review
难度 2/5 1-3 小时 新手友好度 68/100
github/awesome-copilot#5153 · 2 条评论 ·
维护者通常 1 天内回复
-
external-plugin needs-review:MEDIUM ready-for-review
难度 2/5 1-3 小时 新手友好度 68/100
github/awesome-copilot#5149 · 2 条评论 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
github/awesome-copilot#5148 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 82/100
github/awesome-copilot#5147 ·
维护者通常 1 天内回复
查看 github/awesome-copilot 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 83/100
维护者通常 2 天内回复
-
feedback simulation workshop
难度 2/5 1-3 小时 新手友好度 72/100
githubnext/gh-aw-workshop#4755 ·
维护者通常 1 天内回复
-
fix
难度 1/5 1 小时以内 新手友好度 72/100
idean3885/claude-ops-agent#728 ·
维护者通常 1 天内回复
-
deck: add to staging level: easy p-feature: Privacy Policy priority: MUST HAVE ready for product role: front-end size: 0.25pt Waiting on Stakeholder
难度 2/5 1-3 小时 新手友好度 61/100
hackforla/tdm-calculator#3654 · 1 条评论 ·
维护者通常 1 天内回复
-
check:failed feeds:add
难度 1/5 1 小时以内 新手友好度 78/100