recommend

What it does

The skill provides advisory, research-backed architecture recommendations for building or evaluating a knowledge system. It starts from a user-provided use case and constraints, and triggers on specific prompts like "/recommend" or related phrases. It explains the closest preset, reports match quality, and outputs a concrete 8-dimension configuration with rationale and references to research claims. It also outlines an architecture sketch, schema design, processing pattern, and session rhythm, all tied to research-based claims.

How it works

• It reads a set of reference files to configure recommendation behavior, including tradition presets, universal methodology, component blueprints, claim maps, interaction constraints, and topic navigation.
• It determines whether to enter conversational mode or recommendation mode based on the input: empty or a question leads to a clarifying step, while a use-case description proceeds to recommendation.
• It matches the user signals to the closest tradition preset by scoring criteria such as domain, processing style, scale, pain points, and goals.
• It identifies relevant research topics via a claim map and performs semantic searches (using tools like mcp__qmd__deep_search and mcp__qmd__vector_search) to gather 5–15 claims.
• It maps signals to eight configuration dimensions (Granularity, Organization, Linking, Processing, Navigation, Maintenance, Schema, Automation) with confidence levels and rationales.
• It validates configurations against interaction constraints and reports hard blocks or soft warns, including cascades if changing one dimension affects others.
• It outputs a structured recommendation block showing the closest preset, configuration details with confidence, architecture sketch (folder structure, enabled components, skipped components), schema design, processing pattern, session rhythm, trade-offs, and research backing.

When to use it

• When you want an advisory sketch of a knowledge-system architecture without creating files.
• When choosing between presets to guide initial system design based on research claims.
• When you need a justification chain: signal → research claim → recommended dimension.

What it can touch

• The run relies on tools listed in allowed-tools: Read, Grep, Glob, mcp__qmd__search, mcp__qmd__vector_search, mcp__qmd__deep_search, mcp__qmd__get, mcp__qmd__multi_get.
• It reads reference files under CLAUDE_PLUGIN_ROOT (e.g., tradition-presets.md, methodology.md, components.md, claim-map.md, dimension-claim-map.md, interaction-constraints.md).
• It uses mcp__qmd__get / mcp__qmd__multi_get to fetch claims for depth.

Caveats

• This is advisory and does not generate or modify files by itself; it sketches a system for user evaluation.
• The recommendation degrades gracefully if reference files are missing; fewer citations are produced, but structure remains.
• All recommendations must be grounded in stated research claims and the explicit presets; no outcomes beyond the described format are promised.