user_evaluate

User Evaluate

Purpose

Evaluate content through the lens of what Genesis knows about the user. The differentiator vs generic AI summary is the user model — Genesis's accumulated understanding of who this person is, what they care about, and what they're working on. Treat the user's placement of an item as a request for analysis, not as proof that the item is a strong personal fit or should be adopted.

When to Use

• User drops personal content (articles, research, ideas) into the inbox.
• Content is not Genesis-architecture-relevant but matters to the user.
• Inbox monitor classifies an item as user-relevant.
• User invokes /user-evaluate in a foreground session.

Workflow

1. Assemble user context — Read USER.md, query memory_recall MCP for topics related to the content, check user_model_cache and recent observations. USER.md is the floor; the memory system is the ceiling.
2. Fetch content — If the request supplies URLs, fetch every supplied URL and individually address each source; do not stop because the first source seems sufficient. Never evaluate based on URL text alone. Exhaust all access methods before reporting failure.
3. Apply four lenses — Evaluate through all four, do not skip or collapse:
   • What This Is — content-native analysis (argument, evidence, contribution)
   • How This Could Help You — user-model-informed value extraction
   • What We Could Do With It — collaborative actions (Genesis + user)
   • What to Watch — critical assessment (gaps, biases, counterarguments)
4. Tag (report-only) — Suggest Action Timeline (Now/Soon/Someday) and Relevance (Direct/Tangential/Background). These are recommendations, NOT binding metadata.
5. Write output — Structured evaluation in the format below.

Personal-Relevance Evidence Protocol

Find possible value without manufacturing personal fit. For every material claim about relevance, distinguish:

• Explicit — supported by a specific user statement, current goal, active project, known practice, or constraint in USER.md, memory, or observations.
• Inferred — a plausible connection derived from named evidence. Label it as an inference and say what would confirm or falsify it.
• Unknown — no user-specific evidence is available. Give the content-native value and the question it raises; do not backfill a personal story.
• Constraint-conflicted — a stated user boundary (privacy, employer policy, budget, time, platform, or explicit preference) materially limits the value.