Find Editorial Citations

Find sources that genuinely improve the user's work. Treat CiteGuild as a focused member corpus, not a replacement for a broad search engine.

Workflow

1. Extract a focused search query from the topic, passage, claim, or draft. Preserve the user's meaning and avoid sending confidential or unnecessary text.
2. Ask for the user's own domain only when it cannot be inferred and excluding it materially matters. Pass known owned domains as exact values in excluded_domains or repeated --exclude-domain flags.
3. Choose one available transport:
   • For OpenClaw, Hermes, shell agents, or scripts, prefer the citeguild CLI when it is installed.
   • In a client exposing CiteGuild MCP tools, call search_member_articles.
   • Otherwise, call the versioned REST endpoint POST https://citeguild.lvtd.dev/api/v1/search.
4. Start with 10 results. Increase the limit or try one narrower query only when the first pass is insufficient.
5. Rank candidates by direct topical and claim-level relevance. Prefer a smaller set of strong matches over filling a quota.
6. Present each useful candidate with its title, canonical URL, domain, and a short explanation of what it could support.
7. If the user is drafting content, recommend a citation only when the source materially supports the surrounding claim. Inspect the article before representing its contents when the search result alone is insufficient.

All three transports use the same v1 search contract: query is required; limit is 1–50; language is optional; and up to 20 exact excluded domains are allowed.