Unify resource navigation and feature PASK and PIRA

Author: Kaiming-Y

APPLICATIONS.md

@@ -11,6 +11,7 @@ Persistent goals, routines, preferences, and everyday context make assistance us
 | Setting | Starting resources | Proactive decision to inspect |
 |---|---|---|
 | Everyday digital workflows | [Proactive Agent](papers/conference/ICLR2025/proactive-agent-shifting-llm.md); [dot, GAIA, MineContext](PROJECTS.md) | Which unstated task, contextual suggestion, or background update is worth surfacing? |
+| Latent needs and GUI intent recommendation | [PASK](papers/arxiv/2026-04/pask-intent-aware-proactive-agent.md); [PIRA-Bench / PIRF](papers/arxiv/2026-03/pira-bench.md) | When should ongoing conversation or passive screen context produce assistance, an intent recommendation, or silence? |
 | Mobile preferences and routines | [PersonalAlign](papers/conference/ACL2026/personalalign-long-term-gui-intent.md); [FingerTip 20K](papers/conference/ICLR2026/fingertip-20k.md); [KnowU-Bench](papers/arxiv/2026-04/knowu-bench.md) | When does history justify a suggestion, and when should the assistant ask or respect rejection? |
 | Wearable and physical-world assistance | [Satori](papers/conference/CHI2025/satori-proactive-ar-bdi.md); [ProMemAssist](papers/conference/UIST2025/promemassist-working-memory-wearable.md); [WatchGuardian](papers/arxiv/2025-02/watchguardian-personalized-jiti.md) | Does the user's inferred task, cognitive state, or user-defined behavior warrant an intervention? |
 

PERSONAL_AGENTS.md

@@ -32,6 +32,8 @@ The canonical release, licensing, and checked-date entries live in [PROJECTS.md]
 |---|---|---|
 | Desktop and cross-app workflows | [Proactive Agent](papers/conference/ICLR2025/proactive-agent-shifting-llm.md), [ContextAgent](papers/conference/NeurIPS2025/context-agent.md) | [ProAgentBench](papers/arxiv/2026-02/proagentbench.md), [π-Bench](papers/arxiv/2026-05/pi-bench-long-horizon-workflows.md), [VibeLifeBench](papers/arxiv/2026-08/vibelifebench-living-world.md). These are complementary protocols, not necessarily each method's reported benchmark. |
 | Mobile | [PersonalAlign / HIM-Agent](papers/conference/ACL2026/personalalign-long-term-gui-intent.md) | AndroidIntent in PersonalAlign; [FingerTip 20K](papers/conference/ICLR2026/fingertip-20k.md) and [KnowU-Bench](papers/arxiv/2026-04/knowu-bench.md) cover additional suggestion, execution, and interaction questions. |
+| Conversation, meetings, and learning | [PASK / IntentFlow](papers/arxiv/2026-04/pask-intent-aware-proactive-agent.md) | LatentNeeds-Bench tests demand detection and silence in speech-transcript sessions; staged user evidence remains preliminary. |
+| Passive GUI context and intent recommendation | [PIRF / PIRA-Bench](papers/arxiv/2026-03/pira-bench.md) | PIRA-Bench evaluates profile-conditioned future-intent recommendation and noise rejection; PIRF is its memory-aware, state-tracking baseline. Recommendation is distinct from executing the suggested task. |
 | AR glasses and wearables | [Satori](papers/conference/CHI2025/satori-proactive-ar-bdi.md), [ProMemAssist](papers/conference/UIST2025/promemassist-working-memory-wearable.md), [WatchGuardian](papers/arxiv/2025-02/watchguardian-personalized-jiti.md) | Task-specific user studies; a watch's user-defined behavior trigger differs from a learned multimodal dialogue policy. |
 | Continuous egocentric video | [Vinci2 / EgoMemo](papers/arxiv/2026-07/vinci2-egoserve-proactive-video-assistance.md) | EgoServe and [EgoPro-Bench](papers/arxiv/2026-05/egopro-bench-personalized-streaming.md); recorded video or simulated profiles do not establish longitudinal wearable deployment. |
 | Continual embodied collaboration | [PACT](https://arxiv.org/abs/2605.24350) | Cross-day ask/act adaptation in controlled embodied scenarios. Also belongs in the [robot assistance route](APPLICATIONS.md#embodied-and-robot-assistance). |

PROJECTS.md

@@ -29,6 +29,7 @@ This guide covers implemented resources alongside research papers. Entries are g
 |---|---|---|---|---|
 | **Proactive Agent** | Predicts tasks and assistance opportunities from activity/event streams. | Event collection, data generation, agent/reward checkpoints, and evaluation pipeline. | ICLR 2025 method + ProactiveBench; distinguish generated training scenarios from real-user deployment outcomes. | [Paper](https://arxiv.org/abs/2410.12361) · [Repository](https://github.com/thunlp/ProactiveAgent) · [Notes](papers/conference/ICLR2025/proactive-agent-shifting-llm.md) |
 | **ContextAgent** | Uses open-world sensory context to decide on assistance and tool use. | Inference, SFT, tools, prompts, and sensory-context data layout. | NeurIPS 2025 research system; inspect sensing assumptions and the context-to-assistance policy. | [Paper](https://arxiv.org/abs/2505.14668) · [Repository](https://github.com/openaiotlab/ContextAgent) · [Notes](papers/conference/NeurIPS2025/context-agent.md) |
+| **PASK** | IntentFlow detects latent needs and chooses silence, fast intervention, or deeper assistance with hierarchical user/workspace/global memory. | Official repository, project page, system descriptions, and client download links. | Technical report and preliminary user evidence. Public client/documentation links do not establish that every training component or model weight is released; no installation or evaluation was repeated here. | [Paper](https://arxiv.org/abs/2604.08000) · [Repository](https://github.com/xzf-thu/Pask) · [Project](https://xzf-thu.github.io/Pask/) · [Notes](papers/arxiv/2026-04/pask-intent-aware-proactive-agent.md) |
 
 Streaming research implementations have a dedicated [model and framework matrix](STREAMING.md#models-and-frameworks), including JoyAI-VL-Interaction, MOSS-VL-Realtime, OneStreamer, Proact-VL, Response-G1, Eyes Wide Open, MiniCPM-o 4.5, and Gander.
 

README.md

@@ -98,6 +98,7 @@ A layered starting path, selected for **coverage of different mechanisms and evi
     <tr><td>Sensory context → assistance</td><td><a href="https://arxiv.org/abs/2505.14668"><b>ContextAgent</b></a> · <a href="https://github.com/openaiotlab/ContextAgent">Code</a><br><sub>NeurIPS 2025 · context-aware agent</sub></td><td>Move from desktop logs to open-world sensory context and tool-mediated help.</td></tr>
     <tr><td>Persistent agent activation</td><td><a href="https://github.com/openclaw/openclaw"><b>OpenClaw</b></a> · <a href="https://docs.openclaw.ai/gateway/heartbeat">Heartbeat docs</a><br><a href="https://github.com/refixai/proactivity-sdk"><b>Proactivity SDK</b></a><br><sub>Runtime projects · see <a href="PROJECTS.md">Projects &amp; Products</a></sub></td><td>Compare periodic attention checks with durable goals and model-selected wake cadence.</td></tr>
     <tr><td>Product-level initiative</td><td><a href="https://help.openai.com/en/articles/20001530-getting-started-with-your-dot"><b>OpenAI dot</b></a><br><a href="https://github.com/volcengine/MineContext"><b>MineContext</b></a><br><sub>Always-on product / ambient desktop project</sub></td><td>Compare background research and connected-app context with proactive desktop summaries and tips.</td></tr>
+    <tr><td>Latent needs and persistent memory</td><td><a href="https://arxiv.org/abs/2604.08000"><b>PASK / IntentFlow</b></a> · <a href="https://github.com/xzf-thu/Pask">Code / project</a><br><a href="papers/arxiv/2026-04/pask-intent-aware-proactive-agent.md">Notes</a><br><sub>Technical report · need detection and assistance routing</sub></td><td>Connect silent / fast / full assistance decisions to ongoing conversations and hierarchical personal memory.</td></tr>
     <tr><td>Personal routines across devices</td><td><a href="https://aclanthology.org/2026.acl-long.1669/"><b>PersonalAlign / HIM-Agent</b></a><br><a href="PERSONAL_AGENTS.md">Personal Agents guide</a><br><sub>Mobile user history · preference and routine memory</sub></td><td>Separate completing omitted preferences from anticipating an unrequested routine; personal assistance extends beyond desktop interfaces.</td></tr>
     <tr><td>Wearable and physical-world help</td><td><a href="https://arxiv.org/abs/2410.16668"><b>Satori</b></a><br><a href="papers/conference/UIST2025/promemassist-working-memory-wearable.md"><b>ProMemAssist</b></a><br><sub>AR guidance · working-memory-aware intervention</sub></td><td>Ground assistance in inferred user/task state and weigh the benefit of help against interruption.</td></tr>
     <tr><td>Embodied ask / act decisions</td><td><a href="https://arxiv.org/abs/2605.24350"><b>PACT</b></a><br><a href="https://doi.org/10.1109/ICRA55743.2025.11127399"><b>PACE</b></a><br><sub>Continual robot assistance · action-completion timing</sub></td><td>Compare asking versus acting from interaction history with timing physical help from observed task progress.</td></tr>
@@ -106,6 +107,7 @@ A layered starting path, selected for **coverage of different mechanisms and evi
     <tr><td>Full-duplex interaction → agent work</td><td><a href="https://arxiv.org/abs/2604.27393"><b>MiniCPM-o 4.5</b></a> · <a href="https://github.com/OpenBMB/MiniCPM-o">Code</a><br><a href="https://arxiv.org/abs/2609.08977"><b>Gander</b></a> · <a href="https://github.com/Omni-Interaction-Gander/Omni-Interaction-Agent">Code</a><br><sub>Simultaneous perception, speech, and asynchronous work</sub></td><td>Study speech/listening control, scene-driven comments, and the bridge to a background reasoning agent.</td></tr>
     <tr><th colspan="3" align="left">Tier 3 · Evaluation and boundaries — test the route you chose</th></tr>
     <tr><td>Real workflows and personalization</td><td><a href="https://arxiv.org/abs/2602.04482"><b>ProAgentBench</b></a><br><a href="https://arxiv.org/abs/2604.08455"><b>KnowU-Bench</b></a> · <a href="https://github.com/ZJU-REAL/KnowU-Bench">Code</a></td><td>Pair real desktop traces with personalized, consent-aware mobile evaluation.</td></tr>
+    <tr><td>GUI intent recommendation</td><td><a href="https://arxiv.org/abs/2603.08013"><b>PIRA-Bench / PIRF</b></a> · <a href="https://www.pira-bench.top">Project</a><br><a href="https://huggingface.co/datasets/Yuxiang007/PIRA-Bench-data">Data</a><br><sub>Benchmark with a memory-aware recommendation baseline</sub></td><td>Evaluate latent future intents, user profiles, interleaved tasks, and rejection of noisy visual context before instruction-following execution.</td></tr>
     <tr><td>Long-horizon initiative</td><td><a href="https://arxiv.org/abs/2605.14678"><b>π-Bench</b></a> · <a href="https://github.com/Simplified-Reasoning/Pi-Bench">Code</a><br><a href="https://arxiv.org/abs/2608.10875"><b>VibeLifeBench</b></a></td><td>Evaluate hidden intents and act / ask / stay-silent decisions over evolving personal workflows.</td></tr>
     <tr><td>Streaming timing and intent</td><td><a href="https://omnimmi.github.io/"><b>OmniMMI</b></a><br><a href="https://arxiv.org/abs/2512.01707"><b>StreamGaze</b></a> · <a href="https://github.com/daeunni/StreamGaze">Code</a><br><a href="STREAMING.md#evaluation-routes">Streaming evaluation routes</a></td><td>Compare multimodal alert/turn timing with gaze-conditioned intent; inspect proactive subsets separately from general streaming QA.</td></tr>
     <tr><td>Personalized physical-world assistance</td><td><a href="https://arxiv.org/abs/2605.07299"><b>EgoPro-Bench</b></a><br><a href="https://arxiv.org/abs/2607.11523"><b>Vinci2 / EgoServe</b></a><br><sub>Egocentric timing, user context, and silence</sub></td><td>Connect streaming to personalized assistance while keeping recorded-video evaluation separate from live wearable deployment.</td></tr>
@@ -349,7 +351,7 @@ The complete bibliography retains five research areas for archival browsing. Use
 | 2026-02 | **Pushing Forward Pareto Frontiers of Proactive Agents with Behavioral Agentic Optimization** | ICML 2026 | `Optimization` · `Human Factors` · `Safety & Consent` | [![arXiv](https://img.shields.io/badge/arXiv-2602.11351-b31b1b.svg?logo=arxiv&logoColor=white)](https://arxiv.org/abs/2602.11351) [![ICML](https://img.shields.io/badge/ICML-2026-0f766e.svg)](https://icml.cc/virtual/2026/poster/61493) |
 | 2026-03 | **ProEvent: An Event-centric Benchmark for Proactive Agents** | arXiv 2607 / ACL ARR 2026 | `Benchmark` · `Long-horizon` · `Intervention Timing` | [![arXiv](https://img.shields.io/badge/arXiv-2607.17701-b31b1b.svg?logo=arxiv&logoColor=white)](https://arxiv.org/abs/2607.17701) [![OpenReview](https://img.shields.io/badge/OpenReview-Page-8c1b13.svg)](https://openreview.net/forum?id=wypdOy0HrM) [![Notes](https://img.shields.io/badge/Notes-local-64748b.svg)](papers/arxiv/2026-07/proevent-event-centric-benchmark.md) |
 | 2026-04 | **SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization** | arXiv 2604 | `Optimization` · `Skill Learning` · `Memory` | [![arXiv](https://img.shields.io/badge/arXiv-2604.02268-b31b1b.svg?logo=arxiv&logoColor=white)](https://arxiv.org/abs/2604.02268) [![Star](https://img.shields.io/github/stars/ZJU-REAL/SkillZero.svg?style=social&label=Star)](https://github.com/ZJU-REAL/SkillZero) [![Notes](https://img.shields.io/badge/Notes-local-64748b.svg)](papers/arxiv/2026-04/skill0-skill-internalization.md) |
-| 2026-04 | **PASK: Toward Intent-Aware Proactive Agents with Long-Term Memory** | arXiv 2604 | `Long-horizon` · `Memory` · `Intervention Timing` | [![arXiv](https://img.shields.io/badge/arXiv-2604.08000-b31b1b.svg?logo=arxiv&logoColor=white)](https://arxiv.org/abs/2604.08000) [![Website](https://img.shields.io/badge/Website-9cf.svg)](https://xzf-thu.github.io/Pask/) [![Notes](https://img.shields.io/badge/Notes-local-64748b.svg)](papers/arxiv/2026-04/pask-intent-aware-proactive-agent.md) |
+| 2026-04 | **PASK: Toward Intent-Aware Proactive Agents with Long-Term Memory** | arXiv 2604 | `Long-horizon` · `Memory` · `Intervention Timing` | [![arXiv](https://img.shields.io/badge/arXiv-2604.08000-b31b1b.svg?logo=arxiv&logoColor=white)](https://arxiv.org/abs/2604.08000) [![Website](https://img.shields.io/badge/Website-9cf.svg)](https://xzf-thu.github.io/Pask/) [![GitHub](https://img.shields.io/badge/GitHub-Code-181717.svg?logo=github&logoColor=white)](https://github.com/xzf-thu/Pask) [![Notes](https://img.shields.io/badge/Notes-local-64748b.svg)](papers/arxiv/2026-04/pask-intent-aware-proactive-agent.md) |
 | 2026-05 | **CogniFold: Always-On Proactive Memory via Cognitive Folding** | arXiv 2605 | `Memory` · `Intent Inference` · `Benchmark` | [![arXiv](https://img.shields.io/badge/arXiv-2605.13438-b31b1b.svg?logo=arxiv&logoColor=white)](https://arxiv.org/abs/2605.13438) [![HF Paper](https://img.shields.io/badge/HF-Paper-ffcc00.svg?logo=huggingface&logoColor=black)](https://huggingface.co/papers/2605.13438) [![Star](https://img.shields.io/github/stars/OpenNorve/CogniFold.svg?style=social&label=Star)](https://github.com/OpenNorve/CogniFold) [![Dataset](https://img.shields.io/badge/HF-Dataset-ffcc00.svg?logo=huggingface&logoColor=black)](https://huggingface.co/datasets/OpenNorve/CogEval-Bench) [![Notes](https://img.shields.io/badge/Notes-local-64748b.svg)](papers/arxiv/2026-05/cognifold-always-on-proactive-memory.md) |
 | 2026-05 | **MUSE-Autoskill: Self-Evolving Agents via Skill Creation, Memory, Management, and Evaluation** | arXiv 2605 | `Optimization` · `Skill Learning` · `Memory` | [![arXiv](https://img.shields.io/badge/arXiv-2605.27366-b31b1b.svg?logo=arxiv&logoColor=white)](https://arxiv.org/abs/2605.27366) [![Notes](https://img.shields.io/badge/Notes-local-64748b.svg)](papers/arxiv/2026-05/muse-autoskill.md) |
 | 2026-05 | **Do Proactive Agents Need an LLM to Decide When to Act?** | arXiv 2605 | `Intervention Timing` · `Optimization` · `Tool Use` | [![arXiv](https://img.shields.io/badge/arXiv-2605.30152-b31b1b.svg?logo=arxiv&logoColor=white)](https://arxiv.org/abs/2605.30152) [![Notes](https://img.shields.io/badge/Notes-local-64748b.svg)](papers/arxiv/2026-05/do-proactive-agents-need-llm-wake-anchor.md) |

WEBSITE.md

@@ -29,9 +29,9 @@ This validates bibliography coverage, source deduplication, the repository's fiv
 - `BENCHMARKS.md`: structured nine-column evaluation source, searchable resource directory, and selection guide.
 - `STREAMING.md`: model/framework mechanisms, release status, capability routes, and evidence boundaries.
 - `PROJECTS.md`: assistants, implementations, products, and supporting components.
-- `website/content/selected.json`: balanced homepage picks (two agents, two streaming models, two benchmarks), short navigation descriptions, original figure URLs, source attribution, and alternate text.
+- `website/content/selected.json`: homepage picks across agents, models, and benchmarks, including PASK and PIRA-Bench / PIRF, short navigation descriptions, original figure URLs, source attribution, and alternate text.
 
-Homepage selections use three explicit groups: Proactive Agents, Streaming Proactive Models, and Benchmarks, with two picks each. Official overview images retain their source links; a missing image does not exclude a representative work. Colored SVG icons and the clickable collection map provide visual navigation. The Observe → Decide → Assist / wait diagram is an editorial concept, not a reconstructed paper figure.
+Homepage selections use three explicit groups: Proactive Agents, Models, and Benchmarks. PASK adds a latent-need/personal-memory system; PIRA-Bench / PIRF adds GUI intent recommendation and its reference framework. Official overview images retain their source links; a missing image does not exclude a representative work. SVG icons and the clickable collection map provide visual navigation. All directory pages share the same sidebar, with an icon for each entry and a standard current-page marker. Paper, project, and benchmark tags use small outlined boxes. Streaming uses the same outlined tag and `library/?tag=Streaming` filter as every topic. It has no separate metadata badge, top-level navigation item, sidebar item, or exclusive accent color. The existing `/streaming/` guide remains reachable from documentation links for compatibility. Proactive remains the primary visual emphasis. The Observe → Decide → Assist / wait diagram is an editorial concept, not a reconstructed paper figure.
 
 The design adapts the MIT-licensed **Minted Directory Astro** directory template to this repository's static generator. See [website/TEMPLATE.md](website/TEMPLATE.md) for the upstream source, adapted components, and retained license. The website focuses on direct Paper / Repo / Project / Model / Dataset navigation, category browsing, and selected overview figures.