RunningTab: Direct Workspace Interaction with Environment-Side Tabs
RunningTab adds an environment-side task tab so LLM agents track unread files and unmet requirements.
RunningTab is a framework for direct workspace interaction in which an environment-side tab records outstanding requirements, file excerpts with provenance, and listed but unopened candidates. The agent resolves each requirement against matching content or sets it aside, and a finish check surfaces requirements that are still open. Across three benchmarks and three LLMs, it outperforms plain direct workspace interaction and baselines that keep the record in the model. The tab usually retains the values a deliverable needs once those files have been seen.
- Environment keeps a per-task tab of requirements, excerpts, and unopened candidates.
- Finish check returns still-open requirements if the agent tries to complete early.
- Outperforms plain direct workspace interaction and in-model record baselines.
- Validated on three benchmarks with three LLMs.
From the original article
1 min read at arxiv.orgMuch knowledge work produces new deliverables from files a workspace already holds, and LLM agents are beginning to take such work over. Through direct corpus interaction, an agent can search and read any of those files from a terminal with no indexing, and producing a deliverable from many of them in this way is what we call direct workspace interaction (DWI).
By Jinheon Baek · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.