Menu
Find the related work papers in the relatedwork folder based on storyline.md or paper.md content.
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
Detect when the user wants to start a new CoPaper paper project or initialize a workspace, run `copaper --root . init` when needed, then inspect the current workflow phase and immediately begin the appropriate next phase work.
Inspect the current CoPaper workflow phase, move forward, move backward, or jump to a named phase when the user asks to view, advance, return, roll back, or switch phases in OpenCode, then invoke the skill for the active phase. Uses `copaper status --json` and `copaper set-phase` instead of editing state files.
Convert `paper.md` into a final LaTeX paper project using a user-provided conference or journal template placed under `templates/latex/`. Use when the user asks to generate the final LaTeX paper from the CoPaper markdown draft with a local template.
Helps create placeholder data (tables, figures, metrics) for evaluation sections with clear warnings that all numbers must be replaced with real experimental results. Also generates descriptions explaining what the data shows and how it supports the paper's claims and insights.
Detects undefined terms, unclear descriptions, and clarity issues in academic paper drafts to improve readability and understanding.
Teaches agents how to manage a CoPaper project through the OpenCode plugin tools, including dashboard, initialization, artifact readiness, workflow status, event log inspection, and phase updates. Use this skill whenever the task is about workflow/state management rather than direct paper writing.
| name | relatedwork-finder |
| description | Find the related work papers in the relatedwork folder based on storyline.md or paper.md content. |
This skill automatically finds related-work papers. Every substantive step (keyword extraction, Semantic Scholar search, BibTeX sync, PDF download) is delegated to a copaper CLI subcommand, so the skill never calls MCP search tools or spawns subagents. The skill's job is to orchestrate the CLI calls and interact with the user.
| File | Required | When to Read | Purpose |
|---|---|---|---|
storyline.md | Required if present | Step 1 (input to CLI) | Primary source for keyword extraction by copaper relatedwork keywords |
paper.md | Fallback | Step 1 (auto-fallback) | Used by the CLI when storyline.md is missing |
relatedwork/queries.txt | Read/write | Steps 1-2 | One query per line; written by copaper relatedwork keywords, consumed by copaper relatedwork search --queries-file |
relatedwork/search_cache.json | Read/write | Steps 2-3 | Cache for search metadata (paper_id, title, authors, year, venue, bibtex, arxiv_id, pdf_url, source_queries) — written by copaper relatedwork search |
relatedwork/paper_list.bib | Required | Step 3 | BibTeX entries for formalizing the reference list |
| Relatedwork status | Optional | Steps 3-4 | Call copaper_relatedwork_status for read-only catalog, BibTeX, PDF, summary, and cross-index counts |
relatedwork/literature.json | Required after managed import | Steps 2-6 (after import) | Canonical metadata catalog; inspect progress through copaper_relatedwork_status when the plugin tool is available |
.agents/skills/relatedwork-finder/template.md | Required | Step 5 | Template for PDF summary generation; passed to subagent |
Do NOT read writingrules.md — this skill does not need paper structure rules.
copaper relatedwork search CLI, which queries the Semantic Scholar Graph API. S2's index covers arXiv preprints, major CS/AI venues (CVPR, NeurIPS, ICML, ACL, etc.), and most journals. Do NOT call serper_google_search_scholar, websearch_web_search_exa, or any other external search tool from this skill — the CLI is the single source of truth.relatedwork/search_cache.json (envelope: {"papers": [...]}) with one entry per deduplicated paper, including:
paper_id (extracted from S2's BibTeX key, or generated as <author><year><titletoken>)title, authors, year, venuebibtex (sanitized from S2 citationStyles.bibtex)arxiv_id (from S2 externalIds.ArXiv)pdf_url (from openAccessPdf.url, falling back to https://arxiv.org/pdf/<arxiv_id>.pdf)source_queries (which CLI --query arguments returned this paper)tldr (S2 model-generated one-sentence summary, when present — useful for filtering)SEMANTIC_SCHOLAR_API_KEY in the environment before running this skill to lift the shared anonymous rate limit (~100 req / 5 min) up to 1 req/sec. The CLI auto-detects the env var; without it the CLI still works but may hit 429s on busy days (it retries with exponential backoff up to 4 attempts).relatedwork/search_cache.json is reviewed and trimmed by the user, you MUST import it into relatedwork/literature.json via copaper --root . relatedwork import --input relatedwork/search_cache.json. relatedwork/literature.json is the canonical metadata store used by BibTeX sync, downloads, and summary tracking.relatedwork/pdfs/.shi2026streamingvla.pdf).relatedwork/literature.json. Do not fake a success state.Use OpenCode plugin tools for workflow/artifact events when available. The current plugin does not expose a generic tool-call event logger or relatedwork event writer; do not append logs manually in plugin-based workflows.
If the user asks to inspect workflow event history, call copaper_workflow_log for read-only querying of .agents/events.jsonl instead of reading the file directly.
You MUST follow this step-by-step interactive workflow. STOP and wait for user confirmation after each step marked with [WAIT FOR CONFIRMATION].
copaper --root . relatedwork keywords --count 8 (adjust --count if the storyline is unusually broad/narrow). The CLI reads storyline.md (falling back to paper.md), calls the configured LLM via COPAPER_MODEL, and writes the queries one-per-line to relatedwork/queries.txt.cat relatedwork/queries.txt and present the list to the user.relatedwork/queries.txt (add/remove/replace lines) before I run the search?"Run the search CLI consuming the queries file from Step 1:
copaper --root . relatedwork search \
--queries-file relatedwork/queries.txt \
--limit 20 \
--year 2022-2026
默认过滤(可覆盖):--fields-of-study "Computer Science" 默认就生效。要关掉:--fields-of-study ""(空串禁用学科过滤)。
PDF 可用性不做默认筛选 —— 默认保留所有有元数据的论文(无 OA PDF 的依然进 cache)。需要只看可下载的:加 --open-access。
其他常用 flag:--limit N(每条 query,max 100)、--year 2020-2024、--venue "CVPR,NeurIPS"(限定 venue;arXiv.org 也算 venue)。--query "..." 仍接受作为 --queries-file 的替代。
The CLI writes relatedwork/search_cache.json with deduplicated metadata (BibTeX, arXiv ID, PDF URL, venue, TLDR) — you do NOT need to fetch BibTeX, venue, or PDF URLs separately. Do NOT hand-edit the cache to "enrich" it; rerun the CLI with refined queries instead.
ACTION: Read relatedwork/search_cache.json and present a numbered list of papers to the user. Show title, authors[0]+et al., year, venue, and tldr (if present) for each entry.
STOP: Ask "Here is the list of papers the CLI found. Which ones should I keep? (Metadata is already cached for all entries.)"
relatedwork/search_cache.json keeping only the selected papers (preserve the {"papers": [...]} envelope).copaper --root . relatedwork import --input relatedwork/search_cache.json to merge the kept entries into relatedwork/literature.json.paper_list.bib already contains papers missing from relatedwork/literature.json, you MAY rerun copaper --root . relatedwork search --query "<title>" to enrich them, then re-import.copaper --root . relatedwork sync-bib to write metadata-backed entries into relatedwork/paper_list.bib and to import any remaining bib-only entries into relatedwork/literature.json.copaper_relatedwork_status to inspect catalog and BibTeX counts. This tool is read-only; do not replace it with prompt-only status text or a manual .agents/state.json update.relatedwork download or retry tool; do not hand-write download results into JSON. Ask before using any CLI fallback.copaper_relatedwork_status to inspect PDF download status, failures, and summary readiness. This tool is read-only and must not be used to fake download success.relatedwork-summarizer skill to generate summaries now?"