arxiv

name	arxiv
description	arXiv paper discovery — select top 3 significant papers, research via subagents, ingest to wiki, deliver report.
tags	["arxiv","research","paper-discovery","daily"]
triggers	["cron",{"manual":"delegate_task"}]
updated	"2026-05-25T00:00:00.000Z"
created_by	agent

arxiv — ArXiv Research Curator

Discover, select, and research the top 3 arXiv papers or others if the same papers are found or your task requires it. Deliver a summary report with wiki ingestion and cross-links to existing knowledge threads.

See Also

• references/workflow.md — 6-phase discovery and research workflow
• references/patterns.md — MCP→curl fallback, subagent verification, PDF extraction
• templates/research-brief.md — paper selection brief
• templates/report.md — final daily report

Quick Start

1. STEP 0 — Check Shared Inbox (NEW — do this first):

   # Check if there are PDFs in the shared inbox for manual processing
   ls /home/ty/Documents/paper-research/inbox/
   
   # If inbox has items, process them FIRST before fetching new arxiv papers
   # Claim items by adding to research-carryover.md:
   cat /home/ty/Documents/paper-research/research-carryover.md
   # Update claimed_by column, move to processing/, work, move to processed/
   
   # Inbox items are PDFs already downloaded — process as regular papers
   # Set claimed_by = "arxiv-agent" in research-carryover.md
   

2. Layer 2 Load: Read wiki/scratchpad/jobs/sheet.md and wiki/scratchpad/agent-sheets/arxiv/carryover.md

3. Layer 1 Start: Initialize or clear vault.md to act as your episodic scratchpad for this session

4. Run discovery via arXiv API (Phase 1), logging findings to vault.md

5. Select top 3 by significance (Phase 2)

6. Download PDFs via curl — NOT MCP (Phase 3)

7. Delegate research to subagents (Phase 4)

8. Assemble and deliver report (Phase 5)

9. MOP Compression: Compress vault.md into carryover.md (Layer 1 → Layer 2)

Wiki Operations

• Tools: query_knowledge (check existing coverage), wiki_write_page (paper summaries), wiki_update_index (after ingest)
• Constraint: Check wiki with query_knowledge before ingesting — if topic covered, link to existing page instead of duplicating.
• CRITICAL CONSTRAINT: DO NOT interact with the Kanban board or run kanban scripts. Output open items as - [ ] in the ## What Remains section of your carryover.md. The overseer will create Kanban tickets and assign them to you in jobs/sheet.md. Use the standard mcp-project-synapse tools or bash commands (like curl) exclusively.

Critical Paths

• Inbox (check first): /home/ty/Documents/paper-research/inbox/
• Shared carryover: /home/ty/Documents/paper-research/research-carryover.md
• PDF storage: /home/ty/Documents/paper-research/{arxiv_id}v{version}.pdf
• Wiki paper pages: wiki/sources/papers/{slug}.md
• Daily reports: wiki/scratchpad/jobs/reports/arxiv/arxiv-YYYY-MM-DD-top-papers.md
• Carryover: wiki/scratchpad/agent-sheets/arxiv/carryover.md

Inbox Workflow (Step 0)

The inbox holds already-downloaded PDFs that need processing — these bypass the arxiv API fetch step.

1. ls /home/ty/Documents/paper-research/inbox/ — check for items
2. If items exist: claim in research-carryover.md, move to processing/, process like any paper
3. On completion: move to processed/, update carryover with wiki page URL
4. Then proceed with normal daily arxiv discovery if scheduled

FINAL STEP — MOP Compression (Layer 1 → Layer 2)

After completing paper ingestion, read your vault.md (Episodic Trace) and compress it into your strict wiki/scratchpad/agent-sheets/arxiv/carryover.md (Semantic State), adhering to the ~512 token bound. Include:

• Papers processed this cycle (arxiv_id + title + key finding)
• Pages created in wiki/sources/papers/ (slug)
• Open items for next cycle (as - [ ] checklist)
• Last run timestamp

Once compressed, clear or archive your vault.md so the next session starts fresh.

Quality Standards

• Select for significance, not recency
• Wiki pages must cross-link to existing concepts (no orphans)
• 429 from MCP → switch to curl immediately
• Subagent completion must be verified (file exists + grep check)

Kanban Queue

Drain your assigned cards at the start of every cycle:

kanban_list(lane="ready", assignee="<your-profile>")

If the queue is empty, proceed with the agent's normal work. Cross-agent cards (with tenant=, lane=triage) are routed by the Overseer — you do not need to action them directly.

Full contract (call signatures, intents, the tenant trap): wiki/scratchpad/agent-sheets/overseer/references/kanban-coordination.md.