Paper Read — Interactive Paper Reading for OpenCode

Forked from backnotprop/plannotator.
An interactive paper-reading workflow for OpenCode.

What This Is

Given an arXiv ID, this workflow downloads the paper, generates a structured report, opens it in paper-read for multi-round annotation and feedback, then persists the final report and updates a knowledge graph.

Fix: Ask AI Now Works with OpenCode

The original Plannotator's "Ask AI" showed a "Session error" in the OpenCode desktop app. The AI runtime created a new OpenCodeProvider instead of reusing the existing authenticated client. This fork fixes that.

File	Change
packages/ai/types.ts	Added optional client field to OpenCodeConfig
packages/ai/providers/opencode-sdk.ts	Constructor accepts pre-existing client
packages/server/ai-runtime.ts	createAIRuntime accepts opencodeClient option
packages/server/index.ts	Passes opencodeClient to createAIRuntime

Paper Reading Workflow

Flow

1. Download/load paper — arXiv ID, URL, or local PDF/TXT
2. Draft report in memory — markdown, following the Report SOP (do NOT write to disk yet)
3. Call submit_plan — opens paper-read review UI in browser
4. Multi-round annotation — annotate → Request Changes → agent edits → resubmit
5. Approve → write report.md → wiki ingest → build graph

Report SOP

The default report structure follows these 7 items:

1. 论文研究的问题是什么
2. 为什么这个问题值得研究，应用在哪
3. 与其相关工作对比，这个工作特点在哪
4. 技术栈选型是什么
5. 架构流程图是怎么样的
6. 数据流和控制流是怎么做的
7. 结论是什么，有什么 insight

Supported Agents

• OpenCode (only)

Installation

1. Clone this repo

git clone https://github.com/liuforge/paper-read.git

2. Build and install the plugin

cd paper-read
bun install
bun run build

3. Copy the workflow files

cp -r paper-read/.opencode/* /path/to/your/project/.opencode/

4. Add the plugin to your OpenCode config

// .opencode/opencode.jsonc
{
  "$schema": "https://opencode.ai/config.json",
  "plugin": [
    ["@plannotator/opencode@latest", {
      "workflow": "plan-agent",
      "planningAgents": ["plan", "read"]
    }]
  ]
}

5. Restart OpenCode

Usage

/read arxiv:2606.18112

Or switch to read mode and type the arXiv ID directly.

Future Work

• Wiki & knowledge graph persistence — complete the post-approval pipeline: report.md → wiki ingest → build graph for full knowledge base integration
• HTML generation speed — optimize report rendering to reduce latency
• Token efficiency — reduce context usage during paper ingestion and report generation

License

MIT — same as upstream backnotprop/plannotator.