Workspace Search with Citations

This skill makes Ragbot's three-tier memory retrieval usable by the model the way humans use a research library: ask a question, get back passages with the page numbers attached, and answer using those passages — quoting where useful and pointing back to the source every time.

The skill exists because retrieval without citations is indistinguishable from hallucination. A response that says "the deployment uses Cloud Run" is useful only if the reader can verify it. The verification path is the citation: which document, which character range, what does that passage actually say. The model owes the user that path on every factual claim.

When to Apply

• The user asks a question that depends on workspace content (their notes, code, lessons, project context, prior decisions).
• The user asks "where does it say X?" or any variant that explicitly asks for a citation.
• The user asks for a summary or synthesis that draws on multiple documents.
• The user follows up on an earlier answer and the answer needs to stay grounded in the same sources.

When NOT to Apply

• The question is general knowledge that the workspace cannot reasonably answer (math, definitions, public-domain facts). The model answers from training and says so.
• The question is procedural and answered by another skill (e.g., summarize-document when a document is given inline).
• The user explicitly asks for the model's opinion, draft, or generation — in which case draft-and-revise is the right tool. Citations still belong in any factual scaffolding inside the draft.