From paper-wiki
Searches biomedical literature (bioRxiv, medRxiv, PubMed Central) via the paperclip MCP server and optionally folds results into the wiki. Use when the user invokes /paper-wiki:bio-search, asks to "search bioRxiv", "search PubMed", "find biomedical papers on X", "look up a clinical trial on Y", or otherwise wants biomedical-domain literature beyond what arXiv covers.
How this skill is triggered — by the user, by Claude, or both
Slash command
/paper-wiki:bio-searchThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
bio-search is the biomedical-literature surface for paper-wiki. It
bio-search is the biomedical-literature surface for paper-wiki. It
delegates the actual search to paperclip
via the paperclip MCP server (mcp__paperclip__* tools when registered)
and optionally hands hits to /paper-wiki:wiki-ingest so they land in
your wiki the same way arXiv papers do.
This SKILL is opt-in: paperclip is a third-party service with its own auth and tier model. paper-wiki's other SKILLs (digest, analyze, wiki-*) work without it. Only invoke bio-search when the user is specifically asking for biomedical-domain literature.
/paper-wiki:bio-search <query>.analyze/digest thread with "what do
bioRxiv preprints say about this?" or similar.Do not use for:
/paper-wiki:digest with
the daily-arxiv recipe./paper-wiki:analyze./paper-wiki:wiki-query.Check that paperclip MCP is available. Run this exact bash to
call the diagnostics runner via the shim. The export PATH=...
line is mandatory — fresh-install users may not have
~/.local/bin on PATH yet (D-9.34.6).
source "$HOME/.local/lib/paperwiki/bash-helpers.sh" 2>/dev/null || {
echo "ERROR: paper-wiki bash-helpers missing at ~/.local/lib/paperwiki/bash-helpers.sh." >&2
echo " Fix: exit Claude Code and re-open — the SessionStart hook installs the helper." >&2
echo " Persistent failures: ~/.local/lib/ may be unwritable; re-run \$CLAUDE_PLUGIN_ROOT/hooks/ensure-env.sh." >&2
exit 1
}
paperwiki_ensure_path
paperwiki diagnostics
Parse the JSON output and inspect mcp_servers. If paperclip is
missing, stop and show the user the registration steps from
docs/paperclip-setup.md. Do not
run claude mcp add yourself; the user opts in.
As an offline fallback, mention the bundled
recipes/biomedical-weekly.yaml recipe — it shells out to the
paperclip CLI and works without the MCP layer if the user prefers.
Resolve the query. Accept free-text. If the user gives a vague topic ("pathology"), ask one clarifying follow-up before burning a paperclip call: "do you want preprints, peer-reviewed PMC, or both?", "any time window?", etc. Default to bioRxiv + medRxiv + PMC over the last 14 days unless told otherwise.
Call paperclip via MCP. Use the mcp__paperclip__search tool
(or whatever name the registered MCP server exposes). Surface the
raw paperclip results — title, authors, source, abstract, link —
in a numbered list. Do not paraphrase; users want to scan the
actual abstracts.
Offer follow-ups.
/paper-wiki:analyze paperclip:bio_<id> (or arxiv:<id> if the hit has an arXiv id
under external_ids).Wiki/papers/<canonical-id>.md via
MarkdownWikiBackend.upsert_paper directly, then chain to
/paper-wiki:wiki-ingest. (A dedicated paperwiki fetch-pdf
subcommand is a future-Phase candidate; today the wiki backend
handles the upsert without a separate fetcher runner.)Summarize at the end. Tell the user what was added to the wiki
(canonical ids + concept names if any), and suggest the next step:
/paper-wiki:wiki-lint to spot dangling sources, or
/paper-wiki:wiki-query to ask follow-up questions.
| Excuse | Why it's wrong |
|---|---|
| "I'll auto-register paperclip MCP for the user." | Auth is sensitive and the user may be on a metered plan. Hand over the registration command, never run it. |
| "I'll paraphrase the abstracts to save space." | Users want to scan the actual paperclip text — paraphrasing introduces drift and hides what the search engine actually returned. |
| "If paperclip times out, I'll fall back to my training data." | Out-of-date / hallucinated biomedical claims can be dangerous. If the call fails, surface the failure; don't fake the result. |
| "The user said 'biomedical' — I'll just guess they mean PMC." | Different sub-corpora have different freshness and peer-review status. Confirm scope (preprints vs. peer-reviewed) before searching. |
paperclip in mcp_servers and the
user is mid-conversation with high urgency — slow down. Show the
registration steps; do not run claude mcp add.id collides with an existing source under
Wiki/papers/ — surface the duplicate; do not silently overwrite.paperclip appears in the diagnostics mcp_servers field before
the SKILL invokes any MCP tool.Wiki/papers/ if
the user opted to save hits./paper-wiki:wiki-lint surfaces a DANGLING_SOURCE finding for any
newly saved source until the user runs /paper-wiki:wiki-ingest.tool_use_id is logged so the user can audit what
paperclip was asked.npx claudepluginhub kaneyxx/paper-wiki --plugin paper-wikiFetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
Applies a firm's KYC/AML rules grid to parsed onboarding records: assigns risk rating, checks required documents, outputs rule outcomes with citations, and routes for escalation.
Generates daily or weekly digests of activity from connected sources (chat, email, docs, tasks, CRM), highlighting action items, decisions, mentions, and project updates.