Knowledge Synthesis

Knowledge Synthesis is an official Anthropic skill from the open-source knowledge-work-plugins repository. According to the provider, it is the last mile of enterprise search: a compatible Claude agent receives raw results from several connected sources and turns them into a coherent answer with source attribution. The skill is therefore an instructional resource for synthesis, not a connector, search engine, or permission system. Its effectiveness depends on which sources are connected, searchable, current, and authorized for the person asking the question.

From raw hits to a useful answer

The documented workflow combines results from chat, email, knowledge bases, cloud storage, and project trackers. Knowledge Synthesis is intended to cluster related results by topic, rank them for the question, and turn them into a narrative answer. The answer should lead with the finding rather than burying it beneath a description of the search process. Raw results should not be repeated source by source. Instead, overlapping claims are combined, meaningful differences are preserved, and supporting evidence is made visible next to the claim or in a complete source section.

Deduplication with context

The same fact may appear in a chat thread, an email, and an updated design document. The skill identifies signals such as similar text, the same author or sender, nearby timestamps, shared entities, and cross-references between sources. These results can be merged into one narrative item. The most complete version can provide the primary context while unique details from other locations are retained. Similarity does not always mean duplication, however. Different conclusions, competing viewpoints, and meaningful evolution from an earlier to a later version should remain separate and be explained in time.

Attribution and confidence

According to the provider, every claim in the synthesized answer must be attributable to a source. Useful attribution includes the source type, precise location, date or relative time, and when relevant the author, subject, channel, or document title. The skill also assesses confidence using freshness and authority. Freshness matters especially for status questions. For policies and factual questions, maintained knowledge bases and official documents may carry more authority than informal chat messages. Several fresh, authoritative sources agreeing with each other support a direct statement. A single old result, an informal source, or conflicting evidence should lead to an explicit qualification about uncertainty and possible staleness.

Adapting to result size

Knowledge Synthesis describes different levels of detail for small, medium, and large result sets. With only a few results, the agent can present each one with context. With a medium set, it can group evidence by theme and identify the most important sources. With a large set, it should provide a high-level synthesis, state the central findings, identify the source types and result count, and offer to investigate a specific aspect further. This keeps the answer readable without discarding useful nuance or the path back to original evidence.

Boundaries and safe use

The skill does not guarantee factual correctness and does not remove the need for human review. Conflicts must be surfaced rather than silently resolved. Connected content may be confidential or may contain hostile instructions. An agent should treat retrieved text as data, respect authorization boundaries, and never make external changes solely because a search result recommends them. The primary source is the official enterprise-search/skills/knowledge-synthesis path in Anthropic's repository. The surrounding collection is published under Apache-2.0. The Recherche and Content categories describe the documented use case, not a guarantee of completeness, privacy, or security.