Infers user decisions and learns preference rules from approvals and corrections.
npx skills add https://github.com/Jamailar/RedBox --skill wwud
Please help me install this Agent Skill. Skill: wwud Source: https://github.com/Jamailar/RedBox/tree/main/desktop/builtin-skills/wwud Command: npx skills add https://github.com/Jamailar/RedBox --skill wwud
AI assistants often stall at decision points that require user judgment, or make choices that clash with the user's actual taste and operating habits. This Skill models how the current app user would decide, approve, reject, or prioritize work, and turns their corrections into reusable decision rules.
Ask the assistant to invoke the wwud skill to decide which of two draft title options you would pick based on your past choices.
High-intent search queries and answers about installing and using this skill.
It classifies the decision as routine, material, or restricted, then reads the narrowest relevant evidence such as user profiles, creator profiles, session context, and approval history. It returns a structured result with decision, confidence, evidence, risk, and fallback.
It records learning events when the user approves, rejects, or corrects a choice. Each event captures the source, decision, chosen and rejected options, a generalized principle, confidence level, and optional expiry, avoiding turning one-off exceptions into global rules.
No. It may decide routine choices at medium or high confidence and material choices only when high confidence and reversible. Restricted actions like publishing, spending, credential changes, or irreversible deletions always require explicit user confirmation.
It prioritizes the current user message and task metadata, then session context, user and creator profiles, profile bundles, active advisor or member skills, knowledge files, manuscript state, and automation history. Generic personality guessing is avoided when app-local evidence exists.
It should not be treated as authorization for high-risk actions, and it should not blend conflicting user, creator, and advisor role logic into vague compromises. When confidence is low and impact is high, it asks one precise question instead of deciding.
ut0p1anMaintain a knowledge profile to guide automated knowledge-collection decisions.
jonnathan-lsAutomate end-to-end content publishing workflows with a chat-driven 10-stage pipeline.
AtlasRoXAnalyze session history and formulate reusable Instincts from user corrections.
empty-canEvaluate current AI model fit and recommend switching between Opus, Sonnet, or Haiku.
