← ClaudeAtlas

ingestlisted

Batch-ingest documents into a TrialError program's knowledge store — register a source, acquire a document under it, confirm the pre-ingestion cost-estimate gate, and drive the normalize/OCR -> chunk -> embed -> index pipeline to completion. Use this when the user hands you paper(s)/book(s)/rulebook(s)/web pages to add to the corpus, or when resuming a pipeline a prior session left mid-flight.
SentImperior666/TrialError · ★ 0 · AI & Automation · score 70
Install: claude install-skill SentImperior666/TrialError
# /ingest — batch ingestion with a cost-estimate confirm gate Design Section 6 (ingestion pipeline) + Section 12 M7 row. Every stage is idempotent, content-hash-keyed, and resumable via the jobs ledger — killing a worker mid-stage never loses progress, it just needs a `jobs tick` + another `start-worker` to pick back up. 1. **Register the source** (one row per work; sha256-dedups automatically — a duplicate `--content-file` refuses with `dedup_of` pointing at the existing source, no second pipeline run): ``` trialerror ingest add-source --kind paper --title "<title>" \ --license-tier <open|academic_oa|user_owned_scan|commercial_restricted|unknown> \ --acquisition-route <author_posted|institutional|publisher_oa|user_scan|user_delivered|api|web> \ --launch-id <your launch_id> --content-file <path> [--authors "..."] [--year N] [--url ...] ``` License fields are REQUIRED at intake (design's licensing posture, C-0048/49 — never guess a tier; ask the user if unstated). `register` refuses an `--acquisition-route` outside the program's configured allowlist (`trialerror.toml [license]`), when one is configured. 2. **Acquire + enqueue the document** under that source. The raw file MUST resolve under a configured ingest root (default `raw/` or `inbox/` — the "manifest-glob wart" this design named and closed): ``` trialerror ingest add --source-id <SRC-id> --path raw/<file> --launch-id <your launch_id> ``` This prints a *