{"content":"IDBots dev journal — MetaWeb learning roadmap: M4 scope realigned with the shipped knowledge-base feature (commit fa24e317, branch feat/metaweb-learning, docs/metaweb-learning-roadmap.md).\n\nContext: the feat/knowledge-base branch (5 commits) shipped per-bot knowledge bases — registry table + per-KB FTS5/BM25 derived index with document-level incremental learning, knowledge_base_list/query/add_document/learn agent tools, a volatile <knowledge_bases> prompt block, and a nightly [00:00,06:00) auto-learn schedule. Crucially it already closes roadmap gap G6: knowledge_base_add_document accepts MetaWeb pin bodies with sourceType 'metaweb' + pinId provenance (x-kb-source), and the learning-loop prompt gained a step 7 telling bots to save substantial pin bodies into a matching KB.\n\nConsequence: the original M4 design (fetch pins into a raw-docs directory, then metabot-create-wiki absorb/index) is superseded — per-bot knowledge bases are the distillation target, strictly better and already built. M4 is re-scoped to the autonomous trigger + execution layer only.\n\nOwner decisions locked today:\n1. Trigger: owner-assigned study topics only in M4; bot-self-derived topics move to M5, to be derived from the bot's role/persona.\n2. No proactive morning report (multi-bot owners would get spammed). Instead the bot answers learning-status questions truthfully from the jobs record, plus a read-only learning-jobs entry near the knowledge-base UI.\n3. Budget default: 20 pins per topic per night.\n\nM4 scope as rewritten: per-bot study-job queue table (topic/status/processed pinIds/budget/summary, idempotent migration); bot tools to enqueue a topic and to query study status/history; a bounded background cowork session per pending job inside the KB auto-learn window running search_metaweb → read_metaweb_pin → knowledge_base_add_document → knowledge_base_learn at scale (dedupe via processed pinIds + content-hash filename idempotency); a sharpened three-layer memory division in the prompts (full pin bodies → knowledge base, how-to steps → procedure_save, concept facts → knowledge_upsert); UI entry adjacent to the KB tab. Deferred: on-chain publish of distilled snapshots; dream-side procedure consolidation stays parked.\n\nMerge plan: feat/knowledge-base merges to main first (conflict surface confirmed tiny — two mechanical hunks in coworkRunner.ts: learning-loop step 6/7 text and the volatile tail), then main merges back into feat/metaweb-learning where M4 implementation continues.","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}