fux-ingest

What problem does it solve? Turning external documents — web pages, PDFs, Excel files, Word docs, images, and OpenAPI specs — into governed, reviewable rules is manual and error-prone. This Skill batch-ingests many sources at once, drafts each implied rule with provenance, and collects them into a single triage queue so nothing becomes active without human review. ## Core Features & Use Cases - Multi-format batch ingestion: Reads URLs (HTTP or CDP-rendered), PDF, Excel/CSV, Word, TXT/Markdown, images, JSON/YAML, and Swagger/OpenAPI specs, with partial-failure tolerance so one bad source never aborts the batch. - Provenance-tracked drafts: Every drafted rule is forced to status: draft and carries source, source_type, fetched, and source_hash, so fux ingest <id> --recheck later detects source or API contract drift. - Bounded link following and connectors: Optionally discovers documents linked from an HTML page (depth-1, same-origin, extension allow-list, capped) and ingests Jira/Confluence/GitHub items via server-side filtered queries — always list-and-confirm, never auto-active. - Token cost reduction: A deterministic reducer strips boilerplate and keeps rule-bearing sections before drafting, reporting before/after token counts. - Use Case: Point it at an OpenAPI spec and a regulatory PDF; it drafts one rule per endpoint contract and per regulatory clause, flags the regulatory items as verify-against-primary-source, and queues everything for human review and ratification. ## Quick Start Ask the agent to run /fux ingest on your source files or URLs, for example: ingest the OpenAPI spec at ./api/openapi.json and the PDF circular-2026.pdf into draft rules for review.