Menu
Research task classification and model routing guidance
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Maintain and debug the Research Harness Zotero plugin adapter. Trigger on Zotero side panel, Zotero reader, Zotero copy/clipboard, local RH API bridge, PDF attachment, RH-to-Zotero sync, seed-paper previews, plugin restart/dev-refresh, or requests to align Zotero plugin behavior with RH CLI/MCP skills without duplicating existing research workflow skills.
Expand from seed papers by following references and citations inside Research Harness workflows. Trigger on phrases like "citation-trace", "/citation-trace", "引用追踪", "沿引用链扩展", or equivalent requests to grow a paper set from one or more seed papers.
Extract core claims from papers for Research Harness evidence structuring. Trigger on phrases like "claim-extraction", "/claim-extraction", "提取论文声明", "claim extract", or equivalent requests to identify key claims, evidence, assumptions, and limitations from ingested papers.
Decide whether a research stage has enough evidence to advance in Research Harness. Trigger on phrases like "evidence-gating", "/evidence-gating", "证据门控", "现在能推进吗", or equivalent requests to check whether current literature, claims, and artifacts justify moving to the next stage.
Improve Research Harness manuscript figures and visual evidence. Use when a paper/PDF has low-quality figures, placeholder fbox/minipage diagrams, blurry PNGs, weak visual evidence, missing publication-ready graphics, figure/table quality review blockers, or requests to use Codex/ImageGen/GPT Image to generate academic paper figures.
Evaluate and replay final_bundle PDF/LaTeX quality for Research Harness project-level self-evolution. Trigger on phrases like "final bundle quality", "PDF quality replay", "LaTeX quality gate", "AAAI PDF QA", "final_bundle_quality", or equivalent requests to scan publication bundles, classify layout/encoding/recompile defects, and decide whether a skill can be promoted.
| name | task-taxonomy |
| description | Research task classification and model routing guidance |
| public_suite | benchmark |
| task_type | reference |
| data_access_level | local |
| inputs | task description, risk level, expected artifact type |
| outputs | task class, route tier, evaluation guidance |
| gates | choose route by evidence risk, cost, and artifact consequences |
| artifact_schemas | runtime_contract |
| estimated_runtime | 1-5 min |
| allowed-tools | ["Read"] |
Research primitives are organized into 7 categories based on cognitive complexity:
Primitives: paper_search
Characteristics:
Routing: kimi
Primitives: paper_summarize
Characteristics:
Routing: kimi
Primitives: claim_extract, baseline_identify, evidence_link
Characteristics:
Routing: kimi → sonnet fallback
Primitives: gap_detect
Characteristics:
Routing: sonnet → opus fallback
Primitives: hypothesis_gen (planned)
Characteristics:
Routing: opus
Primitives: section_draft
Characteristics:
Routing: sonnet
Primitives: consistency_check
Characteristics:
Routing: opus
The MCP server handles routing internally based on these categories. Claude Code sees only the tool names, not the routing details. This skill is primarily for human reference and debugging.
