dptech-corp/bohrium-skills/bohrium-web-search

Bohrium Web Search

Choose the task

Workflow: find documentation and deliver a sourced answer

1. Identify the product/topic, version if relevant, and what fact the user needs. Include those terms in the query; --limit bounds returned leads, not verification depth.
2. Run a small discovery query:

bohr search "official VASP documentation PAW method" --limit 5 -o json

1. Inspect the actual returned container/URL fields (for example data.organic_results[].link or data.results[].url); carry each selected title, URL and snippet together. Open the selected official page using an available page reader/browser. This Skill's command searches; it does not fetch full pages.
2. Verify that the page supports the requested claim and matches the version/date. Deliver the supported answer with its source URL. If only snippets are available, label the finding as a search lead and state the missing verification.
3. An empty result calls for one more precise query when useful; an authentication/network error calls for fixing that error. Neither supports a claim that the subject does not exist. Stop when the requested fact is supported, rather than accumulating unrelated hits.

Use the free, read-only web-search command for general public web information:

bohr search "official VASP documentation PAW method" --limit 5 -o json
bohr search "official VASP documentation PAW method" --limit 5 -o json -q '.data.organic_results'

The query is one positional string. --limit accepts 1-10 and defaults to 5. Prefer a narrow query containing the subject and desired source type.

Route structured academic or private content elsewhere:

• bohr paper search for paper records and bibliographic filters.
• bohr scholar search for researchers and profiles.
• bohr kb search <kb_id> <query> for content already stored in one of the

user's Bohrium knowledge bases.

Search results are discovery leads, not automatically authoritative facts. Preserve titles, URLs, snippets, and source names from the response; open or verify primary sources when accuracy matters. The current CLI exposes no recency, site/domain, locale, or provider-specific filter.

Raw provider data may include base64 favicons, internal search URLs, related queries, and other extras. JSON is forwarded without normalizing result containers: inspect the actual response first (organic_results[].link and results[].url are supported upstream shapes). The shown -q '.data.organic_results' projection applies only when that field exists; an empty projection is not proof of no hits.

Read `references/commands.md` for the exact command, limit validation, output guidance, and routing boundary.