chore(deps): bump eslint from 9.39.3 to 10.2.0

.agents/plugins/marketplace.json

@@ -0,0 +1,20 @@
+{
+  "name": "innoclaw-local",
+  "interface": {
+    "displayName": "InnoClaw Local Plugins"
+  },
+  "plugins": [
+    {
+      "name": "innoclaw-cli",
+      "source": {
+        "source": "local",
+        "path": "./plugins/innoclaw-cli"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Research"
+    }
+  ]
+}

.claude/skills/paper-discussion-full/SKILL.md

@@ -46,11 +46,13 @@ Produce a comprehensive, evidence-grounded discussion report for one paper by em
 
 - State the core claim, method, setup, and reported results.
 - Separate explicit evidence from inference.
+- When referencing external work for comparison, cite it using `searchArticles` results.
 
 ### Critical Analysis
 
 - Challenge overclaims, weak baselines, missing ablations, and threats to validity.
 - Mark issue severity as `Critical`, `Moderate`, or `Minor`.
+- When identifying missing baselines or comparing to external methods, cite specific papers found via `searchArticles`.
 
 ### Reproducibility Check
 
@@ -80,9 +82,38 @@ End with:
 
 `Overall take: ...`
 
+## 8. References
+
 ## Quality Rules
 
-- Do not fabricate details not present in the paper context.
+- Do not fabricate details not present in the paper context. All external references must come from `searchArticles` results, never from model memory.
 - Keep criticism specific and technically grounded.
 - Preserve nuance rather than collapsing everything into a single score.
 - If paper text cannot be retrieved, state that the report is based on limited context.
+
+## Citation Policy
+
+- Any claim about related work, competing methods, or external benchmarks MUST include a citation retrieved via `searchArticles`. Do NOT cite from memory.
+- When the discussion references methods or results not in the seed paper, call `searchArticles` first and cite what is returned.
+- If no supporting reference is found, mark the claim: **[Unverified — no supporting reference found via search]**.
+- The target paper itself must be cited formally at the top of the report.
+
+### Inline Citation & Reference Format
+
+Use numbered inline citations in the text body (e.g., `[1]`, `[2]`) and collect full references in the `## 8. References` section. All references MUST be rendered in markdown so they are clickable and well-formatted:
+
+**Inline example:**
+> 该方法在 ImageNet 上超越了 ViT **[1]**，但在小数据集上的泛化能力受到质疑 **[2]**。
+
+**References section example:**
+
+```
+## 8. References
+
+1. **Dosovitskiy, A. et al.** (2021). *An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale.* ICLR 2021. [arXiv:2010.11929](https://arxiv.org/abs/2010.11929)
+2. **Liu, Z. et al.** (2021). *Swin Transformer: Hierarchical Vision Transformer using Shifted Windows.* ICCV 2021. [DOI:10.1109/ICCV48922.2021.00986](https://doi.org/10.1109/ICCV48922.2021.00986)
+```
+
+- Each reference line MUST include: **Author(s)** (Year). *Title.* Venue/Journal. Linked DOI or URL when available from `searchArticles`.
+- Use markdown bold for authors, italic for title, and `[text](url)` for clickable DOI/URL links.
+- Number references sequentially as they first appear in the text.

.claude/skills/paper-discussion/SKILL.md

@@ -22,7 +22,7 @@ Use this skill for interactive, grounded question answering about a single resea
 1. Identify the target paper from the user's request.
 2. If the paper is not identifiable from the current context, ask for the paper title, URL, or PDF link before continuing.
 3. If the question requires more than title or abstract knowledge, use `readPaper` proactively to ground the discussion in the paper text.
-4. Use `searchArticles` only when the user asks for related work, comparisons, neighboring methods, or follow-up reading.
+4. Use `searchArticles` when the user asks for related work, comparisons, neighboring methods, or follow-up reading, AND whenever you would otherwise cite an external paper from memory.
 5. Answer the user's actual question directly instead of dumping a generic summary.
 
 ## Discussion Priorities
@@ -46,7 +46,34 @@ Use this skill for interactive, grounded question answering about a single resea
 
 ## Quality Rules
 
-- Do not invent citations, baselines, implementation details, or results.
+- Do not invent citations, baselines, implementation details, or results. Any external citation must be retrieved via `searchArticles` — never recalled from memory.
 - If the abstract is insufficient, say so and read the paper before making specific claims.
 - Do not overstate novelty or significance.
 - If the user asks for comparison, name the comparison axis explicitly: method, benchmark, claim, reproducibility, or application scope.
+
+## Citation Policy
+
+- When citing any paper other than the target paper, the citation MUST come from a `searchArticles` call, not model memory.
+- If the user asks about related work or comparisons, call `searchArticles` before answering and cite returned results.
+- If no reference is found, state: **[Unverified — no supporting reference found via search]** rather than citing from memory.
+- When listing references in a structured answer, collect them in a **References** section at the end.
+
+### Inline Citation & Reference Format
+
+Use numbered inline citations in the text body (e.g., `[1]`, `[2]`) and collect full references at the end. All references MUST be rendered in markdown so they are clickable and well-formatted:
+
+**Inline example:**
+> Transformer 架构已被广泛应用于蛋白质结构预测 **[1]**，并在 CASP14 中取得突破性成果 **[2]**。
+
+**References section example:**
+
+```
+## References
+
+1. **Jumper, J. et al.** (2021). *Highly accurate protein structure prediction with AlphaFold.* Nature, 596, 583–589. [DOI:10.1038/s41586-021-03819-2](https://doi.org/10.1038/s41586-021-03819-2)
+2. **Baek, M. et al.** (2021). *Accurate prediction of protein structures and interactions using a three-track neural network.* Science, 373(6557), 871–876. [DOI:10.1126/science.abj8754](https://doi.org/10.1126/science.abj8754)
+```
+
+- Each reference line MUST include: **Author(s)** (Year). *Title.* Venue/Journal. Linked DOI or URL when available from `searchArticles`.
+- Use markdown bold for authors, italic for title, and `[text](url)` for clickable DOI/URL links.
+- Number references sequentially as they first appear in the text.

.claude/skills/research-ideation-full/SKILL.md

@@ -39,6 +39,7 @@ Turn one seed paper into a structured research ideation report by emulating the
   - novelty
   - connection to the seed paper
   - estimated impact
+  - supporting references (at least 1 per hypothesis, retrieved via `searchArticles`)
 
 ### Feasibility Review
 
@@ -65,6 +66,7 @@ Turn one seed paper into a structured research ideation report by emulating the
   - expected outcomes
   - minimum viable experiment
   - timeline
+  - key references (cite the source for each baseline method and any adopted protocol, retrieved via `searchArticles`)
 
 ### Review
 
@@ -83,10 +85,39 @@ Use this exact structure:
 ## 4. Review Findings
 ## 5. Recommended Actions
 ## 6. Overall Assessment
+## 7. References
 
 ## Quality Rules
 
-- Ground every idea in the seed paper or clearly labeled external evidence.
+- Ground every idea in the seed paper or cited external evidence retrieved via `searchArticles`. Model-memory citations are forbidden.
 - Mark speculation explicitly when it goes beyond the paper.
 - Prefer testable and executable ideas over vague ambition.
 - If paper text is unavailable, say that the ideation is based on limited context.
+
+## Citation Policy
+
+- Every hypothesis rationale, related-work claim, baseline reference, and methodology mention MUST include a citation retrieved via `searchArticles`. Do NOT cite from memory.
+- Use `searchArticles` proactively at the start of each hypothesis to locate supporting and contrasting literature before writing the rationale.
+- If `searchArticles` returns no relevant result for a claim, mark the claim explicitly as: **[Unverified — no supporting reference found via search]**.
+- Do not silently drop unreferenced claims; either find a reference or flag it.
+- The seed paper itself must also be cited formally using the same format.
+
+### Inline Citation & Reference Format
+
+Use numbered inline citations in the text body (e.g., `[1]`, `[2]`) and collect full references in the `## 7. References` section. All references MUST be rendered in markdown so they are clickable and well-formatted:
+
+**Inline example:**
+> 近期研究表明，基于扩散模型的分子生成方法在 drug-likeness 指标上显著优于传统 VAE 方法 **[1]**，同时在合成可达性方面也有改善 **[2]**。
+
+**References section example:**
+
+```
+## 7. References
+
+1. **Xu, M. et al.** (2022). *GeoDiff: A Geometric Diffusion Model for Molecular Conformation Generation.* ICLR 2022. [arXiv:2203.02923](https://arxiv.org/abs/2203.02923)
+2. **Hoogeboom, E. et al.** (2022). *Equivariant Diffusion for Molecule Generation in 3D.* ICML 2022. [arXiv:2203.17003](https://arxiv.org/abs/2203.17003)
+```
+
+- Each reference line MUST include: **Author(s)** (Year). *Title.* Venue/Journal. Linked DOI or URL when available from `searchArticles`.
+- Use markdown bold for authors, italic for title, and `[text](url)` for clickable DOI/URL links.
+- Number references sequentially as they first appear in the text.

.dev.pid

@@ -0,0 +1 @@
+6148

.env.example

@@ -5,9 +5,14 @@ WORKSPACE_ROOTS=D:/Data/research,D:/Data/projects
 # Set this to a LOCAL filesystem path if the project is on a network/shared filesystem (NFS, CIFS, etc.)
 # DATABASE_URL=/tmp/innoclaw/innoclaw.db
 
+# Secret used to sign local authentication session cookies.
+# Set a long random value in production, for example: openssl rand -base64 32
+# AUTH_SECRET=replace-with-a-long-random-secret
+
 # Next.js build directory (optional, defaults to .next in project root)
-# Set this to a LOCAL filesystem path to avoid Turbopack cache errors on network/shared filesystems
-# NEXT_BUILD_DIR=/tmp/innoclaw-next
+# On Next.js 16 / Turbopack this must stay INSIDE the project root (for example .next-local)
+# InnoClaw auto-disables Next's dist-dir lock on lockless network filesystems, so this is optional
+# NEXT_BUILD_DIR=.next-local
 
 # GitHub Personal Access Token (for cloning/pulling private repos)
 GITHUB_TOKEN=ghp_...
@@ -36,6 +41,7 @@ ZHIPU_API_KEY=...
 # SHLAB_INTERN_S1_PRO_BASE_URL=https://your-host/v1
 # SHLAB_INTERN_S1_BASE_URL=https://your-host/v1
 # QWEN_QWEN3_235B_BASE_URL=https://your-host/v1
+# QWEN_QWEN3_5_122B_BASE_URL=https://your-host/v1
 # QWEN_QWEN3_5_397B_BASE_URL=https://your-host/v1
 # MOONSHOT_KIMI_K2_5_BASE_URL=https://your-host/v1
 # DEEPSEEK_DEEPSEEK_V3_2_BASE_URL=https://your-host/v1

.github/workflows/docs.yml

@@ -69,7 +69,7 @@ jobs:
 
       - name: Upload artifact
         if: github.ref == 'refs/heads/main' && github.event_name != 'pull_request'
-        uses: actions/upload-pages-artifact@v4
+        uses: actions/upload-pages-artifact@v5
         with:
           path: docs/_build/html