paper-sync

name	paper-sync
description	Check paper pool health, enrich metadata, download PDFs, and annotate — with relevance filtering to skip noise
public_suite	zotero
task_type	synchronization
data_access_level	local+external
inputs	topic name or topic_id, paper queue, acquisition policy
outputs	coverage report, acquired PDFs, dismissed noise, annotation queue
gates	avoid ingesting noise; keep paper operations topic-scoped
artifact_schemas	paper_pool, topic_paper_notes
estimated_runtime	5-30 min
allowed-tools	["mcp__research-harness__paper_list","mcp__research-harness__paper_coverage_check","mcp__research-harness__paper_acquire","mcp__research-harness__paper_dismiss","mcp__research-harness__topic_show","mcp__research-harness__topic_list","Read"]

Paper Sync

Inspect the paper pool for a topic, enrich missing metadata, download available PDFs, annotate them, and produce a prioritized report of what's still missing. Filters out low-relevance noise before downloading.

When to Use

• After /literature-search or /citation-trace to materialize newly ingested papers
• Before /gap-analysis or /claim-extraction to ensure full-text coverage
• Periodically to check pool health
• When the user asks to "sync papers", "download papers", or "check paper coverage"

Invocation

/paper-sync [topic_name_or_id]

If no topic is specified, list topics and ask which one to sync.

Workflow

Phase 1: Diagnose

1. Call mcp__research-harness__topic_show to get topic details

2. Call mcp__research-harness__paper_list with the topic to get current pool state

3. Classify papers into:

   • Complete: status=annotated, has PDF + structure + annotations
   • Needs download: status=meta_only, has identifier (arxiv_id or doi), relevance != low
   • Needs enrichment: missing title or abstract, has identifier
   • Noise candidates: meta_only, relevance=low, no direct connection to topic focus
   • Unreachable: no identifier, no URL, no PDF

4. Report diagnosis to user:

   Paper Pool Health for "topic-name":
   - Complete: XX papers (XX%)
   - Needs download: XX papers
   - Needs enrichment: XX papers
   - Noise candidates: XX papers (suggest dismiss)
   - Unreachable: XX papers
   

Phase 2: Filter Noise

Before downloading, review noise candidates:

1. For papers with relevance=low AND no annotations AND no PDF:

   • Check if their title/abstract mentions core topic keywords
   • If clearly unrelated, call mcp__research-harness__paper_dismiss with reason
   • If uncertain, keep and let the user decide

2. Report dismissals:

   Dismissed X noise papers: [list with reasons]
   

Phase 3: Acquire

1. Call mcp__research-harness__paper_acquire with topic_id

   • This automatically: enriches metadata → downloads PDFs → annotates → builds unable list

2. Report results:

   Acquisition complete:
   - Enriched: X papers (metadata from S2)
   - Downloaded + Annotated: X papers
   - Failed: X papers
   - Needs manual download: X papers
   

Phase 4: Report Unable-to-Acquire

If there are papers in the unable_to_acquire list:

1. Present them sorted by relevance (high first):

   Unable to acquire (sorted by importance):
   
   HIGH:
   - [paper_id] Title (DOI: xxx) — reason: paywalled
     Hint: https://doi.org/xxx
   
   MEDIUM:
   - [paper_id] Title — reason: no downloadable URL
   
   LOW:
   - (suggest dismissing these)
   

2. For high-relevance papers, suggest:

   • Manual download via institutional access
   • Institutional proxy URLs if applicable
   • Alternative sources (preprint versions, author pages)

3. For low-relevance unable papers, suggest dismissing them

Output

End with a summary table:

Paper Pool Summary (after sync):
| Status     | Count | % |
|------------|-------|---|
| Annotated  | XX    | X |
| Meta-only  | XX    | X |
| Total      | XX    |   |

Coverage: XX% of papers have full-text annotations
Next: /gap-analysis or /claim-extraction

Notes

• paper_acquire respects S2 API rate limits (1 req/s with key)
• PDF annotation uses skip_card=True by default (fast, no LLM for card)
• Papers already annotated are skipped automatically (idempotent)
• Dismissed papers are excluded from future syncs and coverage checks