Routes content to the correct knowledge base using tags, rules, and model classification.
npx skills add https://github.com/AlmogBaku/aos --skill route-almogbaku
Please help me install this Agent Skill. Skill: route Source: https://github.com/AlmogBaku/aos/tree/main/capabilities/kb/skills/route Command: npx skills add https://github.com/AlmogBaku/aos --skill route-almogbaku
kb lint fails any shared base carrying an llm-routed record. - Grant-aware candidate filtering: Only bases where the writing subject holds a route-into grant are considered; with zero candidates the payload is handed back tagged refused and recorded via kb refuse. - Use Case: A capture pipeline receives a note from a chat channel; the Skill matches the channel binding in the registry, writes it to the bound base with kb capture, and stamps kb_routing.method=rule via kb set. ## Quick Start Ask your agent to file this content in the right knowledge base using the route skill, reading the registry at $AOS_REGISTRY to pick the destination.High-intent search queries and answers about installing and using this skill.
Read the kb-registry.yaml for base entries, then resolve in order: explicit user tags, channel bindings and keyword rules, a single model classification call against private bases only, and finally the default base. Stamp the result with kb set using kb_routing.method and kb_routing.status fields.
Rules always run first: channel bindings, then case-insensitive keyword substring matches, with no model call. Model classification runs only if no rule matched and every remaining candidate base is private, accepting results at or above the confidence_bar of 0.7.
No. Model classification is only allowed when every candidate base has audience private, and shared bases accept only explicit-tagged or rule-matched writes. The kb lint command fails any shared base carrying a kb_routing.method of llm.
The payload is handed back to the caller tagged kb_routing.status refused, and a refusal entry is recorded with kb refuse against the explicitly named default base. The content is never silently dropped.
The fallback step writes to the default base with method default and status uncertain when no tag, rule, or confident model match exists. The nightly pass re-classifies these but files better destinations as findings for human review rather than moving them.
kvcopCapture raw ideas into structured draft notes with TL;DR and source tracking.
raosuraj23Analyze losing trades to classify failures and generate knowledge-base entries.
doctordukeRoute tasks to small models and escalate low-confidence tasks to large models.
ai-analyst-labCapture corrections and learnings from user messages during Q&A workflows.