Menu
JSON schema reference for 6 structured output card types used by the Research-Claw dashboard: paper_card, task_card, progress_card, approval_card, file_card, monitor_digest. Read this before outputting any structured card for the first time in a session.
Workspace file management, git versioning, directory conventions, and CLI execution safety rules for the Research-Claw workspace.
Convert Office documents (PPTX, DOCX, XLSX, PDF, HTML, CSV, JSON, XML, images) to Markdown using Microsoft MarkItDown. Provides the agent with conversion strategies for academic and research workflows.
Standard operating procedure for research experiments, data analysis, and visualization. Covers Python/R script execution, statistical analysis, data wrangling (pandas/tidyverse), publication-quality figures (matplotlib/plotly/ggplot2), and experiment reproducibility.
Four-phase iterative protocol for academic writing and document production. Covers outline → draft → self-review → polish with Stop Points, material supplementation loops, version comparison, and mechanical self-check. Includes IMRaD, LaTeX, docx/markdown, citation formatting, and md2pdf.
Standard operating procedure for academic figure generation. Four rendering engines: Python (data viz), Mermaid (flowcharts), AI Image via NanoBanana/OpenRouter (complex diagrams), SVG (vector). Includes engine selection decision tree, ReAct self-correction, NanoBanana configuration, environment detection, and academic style rules.
Standard operating procedure for academic literature search. Covers local library search, 18 academic database APIs (L1), web_fetch direct access (L1.5), browser RPA for CNKI/Google Scholar (L2), and optional API services (L3). Includes domain routing, recency protocol, search iteration protocol (evaluate → refine → re-search up to 3 rounds), and Zotero/EndNote import bridges.
| name | Output Cards |
| description | JSON schema reference for 6 structured output card types used by the Research-Claw dashboard: paper_card, task_card, progress_card, approval_card, file_card, monitor_digest. Read this before outputting any structured card for the first time in a session. |
Use fenced code blocks with the card type as the language tag. Content MUST be
valid JSON — the dashboard parser uses JSON.parse().
ONLY for real academic publications — from API queries, library_search, or
user-identified papers. NEVER for concepts, tools, or non-scholarly content.
Required: type, title, authors (string[]).
Optional: venue, year, doi, url, arxiv_id, abstract_preview,
read_status ("unread"|"reading"|"read"|"reviewed"), library_id, tags.
Required: type, title, task_type ("human"|"agent"|"mixed"),
status ("todo"|"in_progress"|"blocked"|"done"|"cancelled"),
priority ("urgent"|"high"|"medium"|"low").
Optional: id, description, deadline (ISO 8601), related_paper_title,
related_file_path.
Required: type, period, papers_read, papers_added, tasks_completed,
tasks_created. Optional: writing_words, reading_minutes, highlights (max 5).
Required: type, action (string), context (string), risk_level ("low"|"medium"|"high").
Required (for exec approvals): approval_id from exec.approval.requested
— without it, dashboard buttons are non-functional.
Optional: details (must be a JSON object, not a string — e.g. {"paper_count": 7}).
CRITICAL: ONLY copy the file_card from workspace_save or workspace_export tool output verbatim.
NEVER fabricate — causes "file not found" errors.
Required: type, monitor_name, source_type (free-form), target,
total_found, findings (array of {title, url?, summary?}, max 10).
Optional: schedule.
