# AI use cases for Research Workbench Assistant (z-liu-xiugou/research-workbench-assistant) Repo: [z-liu-xiugou/research-workbench-assistant](https://github.com/z-liu-xiugou/research-workbench-assistant) | Use cases: 4 | Page: https://dedupedai.com/use-cases/z-liu-xiugou-research-workbench-assistant/ A local-first research memory tool for Codex that keeps research goals, progress, paper notes, and next steps in your own project folder between AI chat sessions. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Resume thesis work across new Codex chats: https://dedupedai.com/use-cases/z-liu-xiugou-research-workbench-assistant/resume-thesis-work-across-codex-chats/ - Find papers on Crossref and triage them as candidates: https://dedupedai.com/use-cases/z-liu-xiugou-research-workbench-assistant/discover-and-triage-papers-with-crossref/ - Test the record workflow from the command line: https://dedupedai.com/use-cases/z-liu-xiugou-research-workbench-assistant/verify-record-flow-without-codex/ - Review project status in an offline HTML overview: https://dedupedai.com/use-cases/z-liu-xiugou-research-workbench-assistant/review-project-status-in-offline-overview/ ## How to resume thesis work across new Codex chats with Research Workbench Assistant Repo: [z-liu-xiugou/research-workbench-assistant](https://github.com/z-liu-xiugou/research-workbench-assistant) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Save checkpoints and next steps in your project so a new Codex chat can pick up where you stopped. **Who it is for:** A graduate student working on a thesis or paper with Codex, who starts a new chat and needs the previous progress and open questions back. **The problem:** Chat history is not a reliable record of research progress, and re-explaining the project in every new session wastes time and loses decisions. **The outcome:** Progress, evidence locations, open questions, and a resume checkpoint are stored as local files in the project, and a new chat reads the checkpoint before loading only what the current task needs. ### Steps 1. Run `python -X utf8 scripts/install.py` from the cloned repository to copy the research-workbench Skill into your user .agents/skills folder, then restart Codex if the Skill is not recognized. 2. Open your own research project in Codex and send: $research-workbench 请为这个项目初始化科研工作台。先了解我的研究目标，不要编造进展，也不要覆盖已有记录。 3. After a substantial piece of work, send: $research-workbench 请保存这次实质进展、证据位置和下一步，并更新续接断点。还没确认的方案不要写成已确认。 4. In a new chat, send: $research-workbench 继续上次研究。先读当前断点，只加载这次任务需要的资料。 ### Prompt to paste into your agent ```text In the research-workbench-assistant repo, use the research-workbench Skill in the project at to initialize a workbench, save the current progress and next steps, then start a new chat and resume. Expected output: an ACTIVE_CONTEXT resume checkpoint in the project. Verify that unconfirmed plans are not recorded as confirmed. ``` ### Caveat Resume memory is local files the Codex Skill reads, not cloud chat memory, and the README notes that the .gitignore is not a privacy guarantee. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/z-liu-xiugou-research-workbench-assistant/resume-thesis-work-across-codex-chats/ ## How to find papers on Crossref and triage them as candidates with Research Workbench Assistant Repo: [z-liu-xiugou/research-workbench-assistant](https://github.com/z-liu-xiugou/research-workbench-assistant) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Search Crossref by topic and year, save deduplicated candidates, and record keep or exclude reasons. **Who it is for:** A researcher starting a literature review on a new topic who needs a shortlist of papers to screen, without marking them as read. **The problem:** Literature searches produce many results, and it is easy to confuse a search hit with a paper that was actually read. **The outcome:** A candidate queue in CANDIDATES.md and JSON with deduplicated DOIs and search batch records. Candidates stay separate from read notes until you write a note for the same DOI. ### Steps 1. Create a workbench to search in with `python -X utf8 skills/research-workbench/scripts/workbench.py init --root demo-workbench --name "示例科研项目"`. 2. Run `python -X utf8 skills/research-workbench/scripts/workbench.py discover --root demo-workbench --query "renewable energy forecasting" --year-start 2023 --year-end 2025 --limit 10` to query Crossref and save results. 3. Run `python -X utf8 skills/research-workbench/scripts/workbench.py candidates --root demo-workbench` to list the queue as pending, kept, or excluded. 4. Open CANDIDATES.md in the workbench, then write a reason for each keep or exclude decision. Do not mark a candidate as read unless you have actually read it. ### Prompt to paste into your agent ```text In the research-workbench-assistant repo, run the discover command against the workbench at with the query for years 2023 to 2025 and a limit of 10. Expected output: CANDIDATES.md and CANDIDATES.json with pending entries. Verify that no candidate is marked as read. ``` ### Caveat Crossref does not cover every paper, the query returns at most 50 results by relevance, and the search terms are sent to Crossref, so avoid unpublished research details in queries. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/z-liu-xiugou-research-workbench-assistant/discover-and-triage-papers-with-crossref/ ## How to test the record workflow from the command line with Research Workbench Assistant Repo: [z-liu-xiugou/research-workbench-assistant](https://github.com/z-liu-xiugou/research-workbench-assistant) | Difficulty: Intermediate | Time: 15 minutes | Basis: From the README Run the bundled demo to see checkpoint and paper records created, resumed, and audited without Codex. **Who it is for:** A developer or lab tech who wants to confirm the tool works on their machine before using it with Codex on real work. **The problem:** Before trusting a tool with research notes, you need to see how records, resume, and audit behave on sample data. **The outcome:** A demo-workbench folder with an ACTIVE_CONTEXT.md resume file and notes folder built from fictional example records, plus an audit result. ### Steps 1. Open a terminal in the cloned repository root, where demo-workbench must not already exist. 2. Run `python -X utf8 skills/research-workbench/scripts/workbench.py init --root demo-workbench --name "示例科研项目"`. 3. Run `python -X utf8 skills/research-workbench/scripts/workbench.py record --root demo-workbench --input examples/checkpoint.json` and then `python -X utf8 skills/research-workbench/scripts/workbench.py record --root demo-workbench --input examples/paper.json`. 4. Run `python -X utf8 skills/research-workbench/scripts/workbench.py resume --root demo-workbench` and `python -X utf8 skills/research-workbench/scripts/workbench.py audit --root demo-workbench`, then open demo-workbench/ACTIVE_CONTEXT.md and demo-workbench/notes/. ### Prompt to paste into your agent ```text In the research-workbench-assistant repo, run the init, record, resume, and audit commands with --root demo-workbench, using examples/checkpoint.json and examples/paper.json. Expected output: demo-workbench/ACTIVE_CONTEXT.md and a generated note in demo-workbench/notes/. Verify that the audit command completes without errors. ``` ### Caveat The example records are fictional and must not be cited as real literature, and the demo does not use Codex at all. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/z-liu-xiugou-research-workbench-assistant/verify-record-flow-without-codex/ ## How to review project status in an offline HTML overview with Research Workbench Assistant Repo: [z-liu-xiugou/research-workbench-assistant](https://github.com/z-liu-xiugou/research-workbench-assistant) | Difficulty: Beginner | Time: 15 minutes | Basis: From the README Generate a local HTML snapshot of checkpoints, tasks, candidates, and notes for a quick status check. **Who it is for:** A researcher preparing a supervisor meeting or a weekly review who needs one page showing where the project stands. **The problem:** Progress is spread across checkpoints, tasks, candidate papers, and notes, which makes a weekly status check slow. **The outcome:** An OVERVIEW.html file in the workbench that shows the checkpoint, task groups, candidates, and existing notes, with search and links back to the original records. ### Steps 1. Make sure a workbench exists, such as one created by the demo flow in the repository. 2. Run `python -X utf8 skills/research-workbench/scripts/workbench.py overview --root demo-workbench`. 3. Double-click OVERVIEW.html inside demo-workbench to open it in a browser. 4. Before sharing the file, review its contents, since it includes local research information. Run the overview command again after new records are added to refresh it. ### Prompt to paste into your agent ```text In the research-workbench-assistant repo, run the overview command on the workbench at and confirm that OVERVIEW.html was created. Expected output: a static HTML snapshot with no external resource requests. Verify that the page is a snapshot and not a live view. ``` ### Caveat The overview is a snapshot that only updates when the overview command runs again, and the README says to check its contents before sharing because it contains local research information. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/z-liu-xiugou-research-workbench-assistant/review-project-status-in-offline-overview/