Project Progress Desk: Know Where All Your Claude Code and Codex Projects Stand at a Glance

Project Progress Desk: Know Where All Your Claude Code and Codex Projects Stand at a Glance

arXiv:2611.13287v1 [cs.HC] 9 Nov 2026

WANGHAO YE1 , YITING WANG1 , SHWAI HE1 , GUOHENG SUN1 , ZIYAO WANG1 , YEXIAO HE1 , ZHEYU SHEN1 , ZHAOYI LIU1 , BOWEI TIAN1 , DAEUN JUNG1 , ADITYA SARKAR1 , PARASTOO PILEVAR1 , XIAOCHUAN GONG1 , YANCHENG ZHU1 , YINGYING FAN1 , QIONG WANG1 , YUNING ZHANG1 , MENG FENG1 , SIYUAN PENG1 , ZHENLE DUAN1 , SIHAN CHEN2 , ZIYI WANG3 , YIFEI DONG4 , YILONG DAI5 , and ANG LI1,* 1 University of Maryland, College Park, USA 4 University of Washington, USA

2 Carnegie Mellon University, USA

3 Texas A&M University, USA

5 The University of Alabama, USA A3 One screen

A5

A2

Receipts

A6 A7 A1

Agent sessions Claude Code

A4

You

Codex

Fig. 1. Project Progress Desk. A1 Coding agents work across projects and sessions. A2 Receipts report outcomes against each project’s checklist; the collector checks them (✓) and updates progress. Dashed slots lack receipts. A3 The overview, with the blue row enlarged below: a Codex receipt reporting 23 of 30 baseline runs raises this project’s progress from 58.4% to 64.4%. A4 The user responds to pending decisions (purple) in the agent’s chat. A5 Attention counters. A6 Dark bars show receipt-supported progress; light portions retain dated estimates. A7 Status color. Simulated demo data. Codex and Claude Code let developers work on many projects in parallel, but keeping track of them still means switching between chats, terminal sessions and providers. We present Project Progress Desk, a local desktop interface that brings overall progress, blockers and pending decisions into one view. Users can see which projects need attention, inspect what remains, and compare earlier approaches before choosing how to continue. Agent receipts link reported outcomes to each project’s weighted acceptance checklist; the overview distinguishes receipt-supported progress from dated estimates. We describe daily use across 36 projects, including a choice between two implementations based on their recorded outcomes. At the demo, attendees will explore 30 simulated projects, trace completed work to overall progress, and watch a coding agent’s new result update the overview. CCS Concepts: • Human-centered computing → Interactive systems and tools; Visualization systems and tools. Additional Key Words and Phrases: coding agents, AI oversight, human-agent interaction, project overview, dashboards, multitasking * Corresponding author: Ang Li, angliece@umd.edu, University of Maryland, College Park, USA.

Manuscript submitted to ACM

1

2

Ye et al. B1

Simulator + contact model Passed

Baselines on all 10 objects 17 / 30 runs

30% 20%

Paper draft + public release

20%

Outline only

B2

Approaches

Goal 100%

30%

Real-robot transfer check Not started

Σ

Progress & next step

Checklist

58.4% 47+11.4confirmed estimated

B3

58.4%

B4

Learned contact model

30%

Friction randomization

40%

3 / 10 objects; in progress 4 / 10 objects; in progress

B5

0%

Still open: 13 baseline runs, transfer checks and release

Soft contact

100%

Rigid contact

25%

✓ 8 / 8 contact checks passed Failed; 2 / 8 contact checks passed

Fig. 2. Project details before the update in Figure 1 A3 , redrawn from the interface. B1 Item height shows weight, and fill shows the progress fraction. B2 The total combines 47 receipt-supported points with 11.4 from an October 30 agent estimate for transfer and release work. B3 The main marker shows the total, 58.4%. B4 Branches connect approaches to outcomes, including B5 eight passed soft-contact checks. Branch geometry is schematic. Simulated demo data.

1

Introduction

Coding agents can now work on their own for long stretches [12], for example through a whole chip-design flow [24], so developers can run many projects in parallel. Project records are spread across chats, terminal sessions and branches, often across multiple providers and accounts. Switching accounts when a quota runs out adds another place to check. After a few days away, it is hard to recall a project’s progress, the approaches tried, or which agents need an answer. People want to leave agents running while they do other work and review the results afterwards [7, 8]. Knowledge workers face a similar cost when they resume fragmented work [13], which earlier desktop systems addressed by grouping windows and documents by task [5, 17, 22]. CliDeck groups agent sessions by project [11]; Agent Cockpit brings status, approvals and history into one interface [1]. Research interfaces support agent steering [4, 10, 14, 27, 29], execution debugging [6] and trajectory comparison [26]. Jira and GitHub Projects organize issues and workflows. The Desk ties each session’s outcomes to a shared project checklist, so users can see how completed work advances the project’s overall goal. We built Project Progress Desk to answer two questions across about three dozen projects: how far along is each project, and which ones need me? Progress bars and status colors support a quick scan [9]; keeping outcomes and next steps on screen reduces what users must remember [21] (Figure 1 A3 ). We developed the Desk through autobiographical design [15] and will open-source it upon acceptance. This demo contributes: • a project-goal view that brings agent work across tools and sessions together, showing overall progress alongside pending user decisions (Figure 1 A3 – A4 ); • a receipt mechanism that updates progress against a shared acceptance checklist and separates receipt-supported outcomes from estimates (Figure 3 C4 – C5 ); • an approach history that preserves successful and unsuccessful alternatives alongside the remaining project requirements, supporting decisions about how to continue (Figure 2 B1 , B4 ). 2 2.1

Project Progress Desk Overview

The overview uses a compact dashboard layout [20] to fit three dozen projects on a laptop screen, each with a name, percentage and bar (Figure 1 A3 ). Dark bars show progress supported by agent receipts (“Confirmed”); light portions show dated estimates A6 [16, 19, 23]. Status colors identify blockers, pending decisions and missing evidence A7 . Manuscript submitted to ACM

Project Progress Desk

C4

Codex receipt

3

grasp-bench

{ "project_id": "grasp-bench", "items": [{ "id": "grasp-bench-a2", "result": "partial", "measure": {"done": 23, "total": 30, "unit": "runs"}, "evidence": [{"path": "runs/baselines.csv", "sha256": "9f2c41d7...e07a"}] }], "project_update": { "next_step": "Run the last 7 baseline runs." } }

C5

Tactile Grasp Benchmark

C6

At the demo

C1

58.4%

17 / 30 runs

+6 points

C2

30% weight

23 / 30 runs

64.4%

Next step

Run the last 7 baseline runs.

<1 min Codex grasp-bench > baselines: 23/30 > receipt published

C1 C2

✓

C3

Find pending decisions Inspect a blocker Compare approaches

Fig. 3. C4 A Codex receipt links reported work to a project, checklist item, supporting evidence and next step. C5 Six additional baseline runs move overall progress from 58.4% to 64.4%. C6 At the demo: C1 find pending decisions, C2 inspect a blocked project, and C3 compare approaches. Simulated demo data.

Clicking a status counter A5 filters the overview; hovering over a project shows its evidence date, blocker and next step. The header lists each account’s remaining quota. The Desk runs locally as a Windows desktop application. 2.2

Progress from Agent Receipts

The Desk turns work across agent sessions into visible progress toward each project’s goal (Figure 1 A1 ). Each project has an agent-drafted goal and weighted acceptance checklist that users can inspect. Coding agents report outcomes, supporting files, blockers and next steps in receipts (Figure 3 C4 ). After checking a receipt against the checklist and linked files, the Desk updates the overview within a minute (Figure 1 A2 ). Each item’s weight determines how much its completion advances overall progress: 𝑃 =

𝑖 𝑤 𝑖 𝑓𝑖 (Figure 2 B1 – B2 ),

Í

where 𝑤𝑖 is the percentage weight and 𝑓𝑖 the completed fraction. Weights sum to 100%. Fractions come from passed items, counts covering the full item, or retained estimates for unmeasured work. The bar reaches 100% when all checklist items pass and project acceptance is recorded. Figure 3 C4 – C5 shows a receipt increasing completed baseline runs from 17 to 23 out of 30. With baselines weighted at 30%, those six runs add six percentage points to overall progress. Users can open the checklist and supporting files to see how each result contributes to the project. 2.3

Approaches and Evidence

Details of an agent’s work help users understand its behavior [3] and check its output [25, 28], but reviewing them takes effort [2, 8, 18]. In the details view, users can compare earlier approaches (Figure 2 B4 ), inspect the evidence behind project progress, and reopen linked Codex conversations to continue work. The main line B3 keeps overall progress visible while branches B4 show the approaches tried across sessions. Users can compare their outcomes before deciding how to continue. In the example, soft contact has passed all eight checks B5 ; the checklist still shows unfinished baselines, transfer and release work.

Manuscript submitted to ACM

4

Ye et al. D1 36 projects

Receipt-supported

Needs attention · 10

Estimated

No attention flag · 26

100% 50% 0%

D2 Project progress D3 Component checks

30/30 passed

25% D4 Unfinished work

Connect to system Check quality

blocked

Fig. 4. D1 A 36-project workspace on November 3, 2026, with the highlighted project’s D2 overall progress, D3 component checks and D4 unfinished work. Component records are from October 29.

2.4

Daily Use

We began using the Desk in September 2026, and about 30 lab members now use it. Figure 4 D1 shows a 36-project workspace from our daily use, with ten projects needing attention for blockers or pending decisions. We use the Desk to review overnight work and return to projects after time away. Previously, returning to a project meant searching its chats to reconstruct what had been done. The Desk lets us review unfinished requirements and pending decisions, and consult earlier approaches when choosing the next step. In the highlighted hardware project, the Desk’s approach history showed that one implementation met the timing requirement and the other missed it. We chose the former for further integration. The project was at 25% overall progress D2 ; component records showed 30 passed simulation checks D3 , with integration and configuration unfinished and quality testing blocked by storage D4 . 3

Demonstration

In a 3–5-minute session, attendees find projects awaiting their decision C1 , inspect a project blocked by an expired token C2 , compare earlier approaches C3 , and give instructions in linked agent conversations (Figure 1 A4 ). A coding agent then reports new work, and the Desk updates the project’s bar, status and next step within a minute. The Windows demo runs on a laptop and external monitor with 30 simulated projects (Figure 3 C6 ). 4

Discussion and Next Steps

Progress reflects each project’s weighted checklist, with estimated portions distinguished from receipt-supported outcomes. Users can inspect the supporting files to assess reported results. Our observations draw on daily lab use. We plan broader deployment and a controlled comparison with a conventional project table, examining how users identify blocked projects and choose their next task. Acknowledgments We used Codex and Claude to assist with system development and manuscript preparation, and Gemini for manuscript review. The video narration was generated using Chatterbox. We reviewed all content and take responsibility for it.

Manuscript submitted to ACM

Project Progress Desk

5

References [1] Agent Cockpit contributors. 2026. Agent Cockpit: Run Multiple Coding Agents from One Control Room. Open-source software. Accessed November 4, 2026. https://github.com/agent-cockpit/agent-cockpit [2] Ariful Islam Anik and Andrea Bunt. 2026. Designing Effective Training Dataset Explanations: The Impact of Information Depth and Progressive Disclosure. In Proceedings of the 31st International Conference on Intelligent User Interfaces (IUI ’26). ACM, New York, NY, USA, 426–441. doi:10.1145/ 3742413.3789087 [3] Michelle Brachman, Siya Kunde, Sarah Miller, Ana Fucs, Samantha Dempsey, Jamie Jabbour, and Werner Geyer. 2025. Building Appropriate Mental Models: What Users Know and Want to Know about an Agentic AI Chatbot. In Proceedings of the 30th International Conference on Intelligent User Interfaces (IUI ’25). ACM, New York, NY, USA, 247–264. doi:10.1145/3708359.3712071 [4] Ruijia Cheng, Jenny T. Liang, Eldon Schoop, and Jeffrey Nichols. 2026. Mapping the Design Space of User Experience for Computer Use Agents. In Proceedings of the 31st International Conference on Intelligent User Interfaces (IUI ’26). ACM, New York, NY, USA, 646–662. doi:10.1145/3742413.3789132 [5] Anton N. Dragunov, Thomas G. Dietterich, Kevin Johnsrude, Matthew McLaughlin, Lida Li, and Jonathan L. Herlocker. 2005. TaskTracer: A Desktop Environment to Support Multi-tasking Knowledge Workers. In Proceedings of the 10th International Conference on Intelligent User Interfaces (IUI ’05). ACM, New York, NY, USA, 75–82. doi:10.1145/1040830.1040855 [6] Will Epperson, Gagan Bansal, Victor C. Dibia, Adam Fourney, Jack Gerrits, Erkang (Eric) Zhu, and Saleema Amershi. 2025. Interactive Debugging and Steering of Multi-Agent AI Systems. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI ’25). ACM, New York, NY, USA. doi:10.1145/3706598.3713581 [7] Cedric Faas, Sophie Kerstan, Richard Uth, Markus Langer, and Anna Maria Feit. 2026. Design Considerations for Human Oversight of AI: Insights from Co-Design Workshops and Work Design Theory. In Proceedings of the 31st International Conference on Intelligent User Interfaces (IUI ’26). ACM, New York, NY, USA, 804–821. doi:10.1145/3742413.3789100 [8] Madeleine Grunde-McLaughlin, Hussein Mozannar, Maya Murad, Jingya Chen, Saleema Amershi, and Adam Fourney. 2026. Overseeing Agents Without Constant Oversight: Challenges and Opportunities. arXiv:2602.16844. https://arxiv.org/abs/2602.16844 [9] Christopher G. Healey and James T. Enns. 2012. Attention and Visual Memory in Visualization and Computer Graphics. IEEE Transactions on Visualization and Computer Graphics 18, 7 (2012), 1170–1188. doi:10.1109/TVCG.2011.127 [10] Brian Kitano, Evan Carlson, Bryan Russett, and Alex Kesling. 2026. Managing Multi-Agent Research Systems: A Dashboard for Human Oversight of Coordinating AI Agents. In CHI 2026 Workshop on Human-centered Evaluation and Auditing of Language Models (HEAL@CHI ’26). https://healworkshop.github.io/#accepted [11] Or Kuntzman. 2026. clideck: A Local Workspace for CLI Agents. Open-source software. Accessed November 4, 2026. https://github.com/rustykuntz/ clideck [12] Thomas Kwa, Ben West, Joel Becker, Amy Deng, Katharyn Garcia, Max Hasin, Sami Jawhar, Megan Kinniment, Nate Rush, Sydney Von Arx, Ryan Bloom, Thomas Broadley, Haoxing Du, Brian Goodrich, Nikola Jurkovic, Luke Harold Miles, Seraphina Nix, Tao Lin, Chris Painter, Neev Parikh, David Rein, Lucas Jun Koba Sato, Hjalmar Wijk, Daniel M. Ziegler, Elizabeth Barnes, and Lawrence Chan. 2025. Measuring AI Ability to Complete Long Software Tasks. arXiv:2503.14499. https://arxiv.org/abs/2503.14499 [13] Gloria Mark, Victor M. Gonzalez, and Justin Harris. 2005. No Task Left Behind? Examining the Nature of Fragmented Work. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’05). ACM, New York, NY, USA, 321–330. doi:10.1145/1054972.1055017 [14] Hussein Mozannar, Gagan Bansal, Cheng Tan, Adam Fourney, Victor Dibia, Jingya Chen, Jack Gerrits, Tyler Payne, Matheus Kunzler Maldaner, Madeleine Grunde-McLaughlin, Eric Zhu, Griffin Bassman, Jacob Alber, Peter Chang, Ricky Loynd, Friederike Niedtner, Ece Kamar, Maya Murad, Rafah Hosn, and Saleema Amershi. 2025. Magentic-UI: Towards Human-in-the-loop Agentic Systems. arXiv:2507.22358. https://arxiv.org/abs/2507.22358 [15] Carman Neustaedter and Phoebe Sengers. 2012. Autobiographical Design in HCI Research: Designing and Learning through Use-It-Yourself. In Proceedings of the Designing Interactive Systems Conference (DIS ’12). ACM, New York, NY, USA, 514–523. doi:10.1145/2317956.2318034 [16] Mahsan Nourani, Lien Nguyen, Carey Barry, Qingchu Jin, and Melanie Tory. 2026. The HEART Interface: Visualizing Risk Score Uncertainty in the Cardiothoracic ICU. In Proceedings of the 31st International Conference on Intelligent User Interfaces (IUI ’26). ACM, New York, NY, USA, 1019–1036. doi:10.1145/3742413.3789109 [17] Nuria Oliver, Greg Smith, Chintan Thakkar, and Arun C. Surendran. 2006. SWISH: Semantic Analysis of Window Titles and Switching History. In Proceedings of the 11th International Conference on Intelligent User Interfaces (IUI ’06). ACM, New York, NY, USA, 194–201. doi:10.1145/1111449.1111492 [18] Saumya Pareek, Jarod Govers, Naja Kathrine Kollerup, Emily Wong, Eduardo Velloso, and Jorge Goncalves. 2026. Sensemaking in Multi-Agent LLM Interfaces: How Users Interpret Transparency and Trustworthiness Cues. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI ’26). ACM, New York, NY, USA. doi:10.1145/3772318.3791157 [19] Snehal Prabhudesai, Leyao Yang, Sumit Asthana, Xun Huan, Q. Vera Liao, and Nikola Banovic. 2023. Understanding Uncertainty: How Lay Decision-makers Perceive and Interpret Uncertainty in Human-AI Decision Making. In Proceedings of the 28th International Conference on Intelligent User Interfaces (IUI ’23). ACM, New York, NY, USA, 379–396. doi:10.1145/3581641.3584033 [20] Alper Sarikaya, Michael Correll, Lyn Bartram, Melanie Tory, and Danyel Fisher. 2019. What Do We Talk About When We Talk About Dashboards? IEEE Transactions on Visualization and Computer Graphics 25, 1 (2019), 682–692. doi:10.1109/TVCG.2018.2864903 [21] Mike Scaife and Yvonne Rogers. 1996. External Cognition: How Do Graphical Representations Work? International Journal of Human-Computer Studies 45, 2 (1996), 185–213. doi:10.1006/ijhc.1996.0048 Manuscript submitted to ACM

6

Ye et al.

[22] Jianqiang Shen, Jed Irvine, Xinlong Bao, Michael Goodman, Stephen Kolibaba, Anh Tran, Fredric Carl, Brenton Kirschner, Simone Stumpf, and Thomas G. Dietterich. 2009. Detecting and Correcting User Activity Switches: Algorithms and Interfaces. In Proceedings of the 14th International Conference on Intelligent User Interfaces (IUI ’09). ACM, New York, NY, USA, 117–126. doi:10.1145/1502650.1502670 [23] Ian Thomas, Song Young Oh, and Danielle Albers Szafir. 2024. Assessing User Trust in Active Learning Systems: Insights from Query Policy and Uncertainty Visualization. In Proceedings of the 29th International Conference on Intelligent User Interfaces (IUI ’24). ACM, New York, NY, USA, 772–786. doi:10.1145/3640543.3645207 [24] Yiting Wang, Wanghao Ye, Yexiao He, Yiran Chen, Gang Qu, and Ang Li. 2025. MCP4EDA: LLM-Powered Model Context Protocol RTL-to-GDSII Automation with Backend Aware Synthesis Optimization. arXiv:2507.19570. https://arxiv.org/abs/2507.19570 [25] Yang Wu, Yao Wan, Mennatallah El-Assady, and April Yi Wang. 2026. StepMIND: A Visual Framework for Stepwise, Multimodal, and Bidirectional Explanations of AI-Generated Data Analysis Pipeline. In Proceedings of the 31st International Conference on Intelligent User Interfaces (IUI ’26). ACM, New York, NY, USA, 522–539. doi:10.1145/3742413.3789070 [26] Yukun Yang, Simret Araya Gebreegziabher, Hojun Yoo, Charles Chiang, Chaoran Chen, Annalisa Szymanski, Hyo Jin Do, Zahra Ashktorab, Werner Geyer, Diego Gómez-Zará, and Toby Jia-Jun Li. 2026. EvalAgent: Interactive Comparative Evaluation of Computer-Using GUI Agents. In Companion Proceedings of the 31st International Conference on Intelligent User Interfaces (IUI Companion ’26). ACM, New York, NY, USA, 171–175. doi:10.1145/3742414.3795096 [27] Xuan Zhao, Jiwoong Sohn, Qinyue Zheng, and Michael Moor. 2026. AgentGUI: An Interface for Observing and Steering Long-Running AI Agents. arXiv:2607.26300. https://arxiv.org/abs/2607.26300 [28] Runtao Zhou, Giang Nguyen, Nikita Kharya, Anh Nguyen, and Chirag Agarwal. 2026. Improving Human Verification of LLM Reasoning through Interactive Explanation Interfaces. In Proceedings of the 31st International Conference on Intelligent User Interfaces (IUI ’26). ACM, New York, NY, USA, 456–473. doi:10.1145/3742413.3789134 [29] Haomin Zhuang, Hanwen Xing, and Xiangliang Zhang. 2026. AgentClick: A Skill-Based Human-in-the-Loop Review Layer for Terminal AI Agents. In Proceedings of the ACM Conference on AI and Agentic Systems (CAIS ’26). ACM, New York, NY, USA, 1372–1378. doi:10.1145/3786335.3813232

Manuscript submitted to ACM

Related documents

• LabOSBench: Benchmarking Computer Use Agents for Scientific Instrument Control#280226
• The Diffusion of the Codex#739076
• LLMVul: A Vulnerability-Labeled Dataset of LLM-Generated C/C++ Functions from Real Production Repositories#998519
• Resolución UNRN AND CDEyVE Nº 35/2026. Aprobar el Protocolo para la realización de la Práctica Profesional Supervisada de la Carrera de Ingeniería en Computación de la Sede Andina.#1461293
• LLMs for research: quality, speed and cost. Comparison with Pi, replaying project tickets from Imperial Dragon Harness workflows#1519456