Alphalab Ustc Researchclaw Researchclaw

Research Taste Profile

⚙️ Step 0 — Read the Research Profile (Always First)

Before running any capability, load the user's research profile.

Location: ~/.openclaw/workspace/research-claw-config.md

If this file does not exist, use these defaults silently and mention at the end:

💡 想定制推荐兴趣？试试说「更新我的研究画像」

# Default profile (used when no config found)
research_direction: "Large language models, reinforcement learning, agentic AI"
seed_papers: []
keywords:
  - large language models
  - reinforcement learning
  - agentic AI / AI agents
  - retrieval-augmented generation
  - multimodal models
whitelist_authors: []
learned_preferences:
  accept: []
  reject: []

Config fields reference:

• research_direction — free-text description of the user's research focus
• seed_papers — list of arXiv IDs the user considers gold-standard references
• keywords — interest topics used for Paper Scout search queries
• whitelist_authors — researcher names to prioritize in recommendations
• learned_preferences.accept — keywords/topics user has explicitly liked
• learned_preferences.reject — keywords/topics user has skipped or disliked

Goal: Maintain the user's research preference profile and render it as a visual HTML page.

Triggers: 更新我的研究画像 · 我的研究画像 · research profile · auto-learn (from Paper Scout feedback)

Config file format

Location: ~/.openclaw/workspace/research-claw-config.md

# ResearchClaw Config
# Auto-maintained by the agent. You can also edit manually.

research_direction: >
  PhD researcher in large reasoning models and agentic memory systems.
  Focus on RL-based training, long-context reasoning, and retrieval-augmented agents.

seed_papers:
  - 2503.19823   # AutoRefine
  - 2412.XXXXX   # MemOCR
  - 2502.XXXXX   # ReMemR1

keywords:
  - large language models
  - reinforcement learning
  - agentic memory
  - long-context reasoning
  - retrieval-augmented generation
  - multimodal agents

whitelist_authors:
  - Yaorui Shi
  - An Zhang
  - Xiang Wang

learned_preferences:
  accept:
    - RL-based reasoning
    - verifiable rewards
    - memory augmentation
  reject:
    - pure NLP classification
    - computer vision only
    - medical imaging

topic_stats:
  - topic: "Reinforcement Learning"    count: 12   pct: 35
  - topic: "LLM Reasoning"             count: 9    pct: 26
  - topic: "Agentic AI"                count: 7    pct: 20
  - topic: "Multimodal"                count: 4    pct: 12
  - topic: "RAG"                       count: 2    pct: 7

Operations

View profile (我的研究画像):

• Load config, regenerate HTML visual profile (see HTML step below)
• In chat, show a compact text summary:

  🧠 你的研究画像
  方向: {research_direction (first sentence)}
  关键词: {keywords joined by · }
  种子论文: {N} 篇
  关注作者: {whitelist_authors joined by , }
  偏好: +{accept topics} / −{reject topics}
  

Update profile (更新我的研究画像 [any description]):

1. Ask the user (or parse from their message) for updated:
   • Research direction description
   • New keywords to add/remove
   • New seed paper IDs to add
   • New author names to whitelist
2. Update the config file
3. Regenerate HTML profile
4. Confirm in chat

Auto-learn (triggered by feedback on Paper Scout results):

1. Parse the user's feedback signal
2. Update learned_preferences.accept or .reject
3. Update topic_stats by incrementing count for relevant topics
4. Save config silently (no need for separate confirmation, just note briefly)

Generate HTML profile

Template location: {SKILL_DIR}/templates/research-profile.html

1. Load with read
2. Replace placeholders:

3. Save to ~/.openclaw/workspace/research-claw-output/research-profile.html
4. Report: 🧠 研究画像已更新 → ~/.openclaw/workspace/research-claw-output/research-profile.html

⚠️ Error Handling

🛠️ HTML Template Usage — General Guide

This section applies to Capabilities 2, 3, and 4.

Finding the skill directory

The skill directory (where templates live) is the folder containing this SKILL.md file. Typical path: ~/.openclaw/skills/research-claw/ Templates are at: ~/.openclaw/skills/research-claw/templates/

If you cannot determine the skill directory, use exec to find it:

find ~/.openclaw/skills -name "paper-note.html" 2>/dev/null | head -1

Output directory

Default: ~/.openclaw/workspace/research-claw-output/

Create if needed:

mkdir -p ~/.openclaw/workspace/research-claw-output

The user can override the output directory by setting output_dir in their config.

Filling placeholders

1. Load template with read tool
2. In your reasoning, create a complete mapping of {{PLACEHOLDER}} → value
3. Perform a full string replacement for every placeholder
4. If a placeholder has no content (e.g., no code URL), use a sensible default:
   • URLs: #
   • Text: N/A or an empty string
   • Counts: 0
5. Write the result with write tool

Never leave unfilled {{PLACEHOLDER}} tags in the output HTML.

Source: AlphaLab-USTC/ResearchClaw — distributed by TomeVault.