github / github/copilot-plugins

Proposal: include OrcaReplay (MCP server for recording and replaying a Copilot session)

未關閉
#91 0 則留言 0 個 reaction 已指派 0 人 在 GitHub 檢視
主要語言
PowerShell
星號
360
分支
125
平均合併
5 天 17 小時
30 天內合併 PR
2

描述

Proposing [OrcaReplay](https://github.com/Continuum-AI-Corp/OrcaReplay) for the collection, following the shape of #29.

**Disclosure: I maintain it.** Apache-2.0, npm, Node 20+, no paid tier, no account, no hosted service.

## What it is

It records a coding-agent session at the HTTP boundary to the model provider — from outside the process — and can serve that recording back so the same session runs again with the provider unreachable and no key. The recording is also readable: which turn changed a file, which command ran, where the build broke, answered from the trace rather than from the agent's own summary of its context window.

## What it would contribute here

An **MCP server** over stdio with six tools — `orca_list`, `orca_show_run`, `orca_graph`, `orca_checkpoints`, `orca_replay`, `orca_compare`. Four are read-only over a local trace store. There is also a skill (`agent-run-forensics`) that drives them with a fixed protocol, if a skill entry is the better shape.

Nothing is installed into Copilot: the recorder wraps the process it launches and moves a base-URL variable for that child only.

## Why I am asking rather than opening a PR

The **External Plugins** list is currently five entries, all GitHub or Microsoft. I could not find written criteria for community submissions, and #35 suggests I am not the only one unsure where the line sits between this repo, `awesome-copilot`, and self-publishing.

So: **is a third-party plugin in scope here at all?** If yes, I will open the PR in whatever shape you prefer — an External Plugins entry, a skill under `plugin/`, or an MCP entry once that section lands. If no, that is a useful answer too and I will take it to `awesome-copilot` instead; no follow-up needed.

## Three limits, stated up front rather than found later

- **A matching replay is not a determinism result.** It shows the recorded run reproduces, not that the model is stable across calls.
- **`egress=blocked` means model-provider egress only.** Recorded tool calls still execute for real on replay — a recorded `curl` reaches the network. Replay is not a sandbox.
- **Embedding calls are not captured** by the default adapter, so a retrieval step is absent from the trace even when the chat turns are complete.

Happy to answer anything else, and equally happy to hear this is out of scope.

貢獻指南

開啟貢獻指南

研究方向

先查看目前的 External Plugins 清單以及 #29 和 #35 中的討論,然後比較此處描述的 MCP-server 和 skill 提議形式。當收到關於第三方外掛是否在範圍內,以及如果在範圍內應優先採用哪種貢獻形式的決定後,即視為完成。

由索引模型根據 Issue 內容生成。

評估

技術堆疊
node.js
領域
tooling
Issue 類型
功能
難度
5/5
預估耗時
一週以上
活躍度
活躍
描述清晰度
需要釐清
新手友好度
25/100

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。