What problem does it solve? Bulk-ingesting large data sources (audio libraries, email takeouts, document corpora, chat exports) into a knowledge brain fails when progress lives in agent memory, trials are skipped, or crashes force restarts from scratch. This Skill provides a full lifecycle discipline so large ingestion jobs survive crashes, session boundaries, and multi-worker fan-out. ## Core Features & Use Cases - Ten-phase lifecycle: SCHEMA → ACCESS → TRIAL → EVALUATE → IMPROVE → CODIFY → TEST → SKILLIFY → BULK → MONITOR, with a mandatory 5-10 diverse trial loop before any bulk run. - Durable manifest state: a manifest.json source of truth (plus rendered MANIFEST.md) built from authoritative source enumeration, with status derived from artifacts on disk rather than assertions. - Idempotent multi-worker execution: disjoint shard partitioning, checkpointing, per-worker progress files, and verification of outputs on disk instead of trusting subagent success claims. - Use Case: Ingest a 400K-message email takeout into brain pages by defining a page schema, trialing 10 diverse threads, codifying the deterministic transform, then fanning out sharded workers whose progress is tracked and resumable via the manifest. ## Quick Start Ask the agent to bulk ingest all of a chosen data source into the brain and have it build a resumable manifest before processing anything.
