Commit Graph

5 Commits

Author SHA1 Message Date
Max Loffgren 1526cf1d4b Move developer and research index skills out of the build repo.
Those are session skills; this repo is for wiring Firecrawl into product code.
2026-08-20 11:39:18 -07:00
Himadri Mishra 798926666a fix: deliver the research-index skill and describe it accurately (#10)
* fix(discoverability): register index skills and name the paper corpus

The research and developer index skills existed on disk but were absent
from the Claude plugin manifest, so Claude Code never loaded them. No
file in the repo carried any biomedical vocabulary, and nothing warned
that `categories: ["research"]` on /search is a website filter rather
than the paper index.

- .claude-plugin/plugin.json: add firecrawl-research-index and
  firecrawl-developer-index to the skills array (both dirs already
  existed; .codex-plugin globs ./skills/ and already picked them up)
- .cursor-plugin/plugin.json: add "skills": "./skills/" so Cursor
  resolves skills at all
- firecrawl-research-index: name the corpus (PubMed, bioRxiv, medRxiv,
  arXiv) in the description and in a new "What is in the index"
  section; add a disambiguation entry for the categories: ["research"]
  website filter, mirroring the pattern the developer-index skill
  already uses for its own categories: ["developer"] twin
- firecrawl-build-search: escalate paper search and developer questions
  to the index skills instead of /search
- firecrawl-build + references/endpoint-selection.md: route to both
  index skills alongside /scrape, /search, /interact
- README.md, AGENTS.md: list the two index skills

Also fixes pre-existing invalid YAML frontmatter in
skills/firecrawl-build/SKILL.md, where an unquoted ": " inside the
description made the whole block unparseable by a strict YAML reader.

* fix(skills): describe the research category filter accurately

The `categories: ["research"]` site list explicitly includes
pubmed.ncbi.nlm.nih.gov, biorxiv.org, and medrxiv.org, so saying the
filter "does not reach" those sources was wrong. It does reach their web
pages; what it does not do is query the paper records in the research
paper index — no abstract search, related-paper or citation-graph
expansion, canonical paper metadata, or in-body passages.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-12 21:10:08 +05:30
Eric Ciarla 67060ec3e8 Focus on search, scrape, and interact. 2026-04-11 14:01:50 -04:00
Eric Ciarla 6569b2acac Fix naming collisions. 2026-04-08 17:00:41 -04:00
Eric Ciarla 55733408b4 Added Firecrawl Skills 2026-04-08 16:39:18 -04:00