Ingest entire blogs, newsletters, and RSS/Atom feeds into deduplicated brain source pages.
npx skills add https://github.com/garrytan/gbrain --skill blog-ingest
Please help me install this Agent Skill. Skill: blog-ingest Source: https://github.com/garrytan/gbrain/tree/main/plugin/skills/blog-ingest Command: npx skills add https://github.com/garrytan/gbrain --skill blog-ingest
Manually saving every post from a blog or newsletter archive into a knowledge base is slow, error-prone, and produces duplicates on re-runs. This Skill automates whole-publication ingestion: it discovers the feed, walks the full archive, normalizes each post, deduplicates by canonical URL, and writes clean source pages that are safe to re-run.
sources/articles/<publication-slug>/, then hands the batch to the brain-ingest-gate skill for entity and backlink enrichment.Ask the agent to ingest this whole blog into the brain, walking its feed and writing one deduplicated page per public post.
High-intent search queries and answers about installing and using this skill.
Provide the publication URL and the skill discovers its RSS/Atom feed or sitemap, enumerates the full archive, then writes one normalized Markdown page per public post. It tests on 3-5 posts first, then ramps to the full archive with pacing between fetches.
Each post is keyed by its canonical URL after stripping tracking parameters and resolving redirects. Before writing, the brain is searched for that URL; existing pages are skipped, making re-runs idempotent and cheap.
No. Gated posts are detected via paywall markers or truncated preview bodies and skipped with a logged reason. The skill never uses alternate endpoints, session cookies, or credentials to reach gated content.
blog-ingest handles publication scope: whole blogs, feeds, and archives requiring enumeration. idea-ingest handles single-item scope: one article or essay URL in, one page out. Single URLs are routed away from blog-ingest automatically.
Pages under roughly 50 words or matching subscribe/paywall boilerplate are flagged as husks after the run. Each husk is re-fetched slowly; real content rewrites the page, while confirmed gated posts are deleted and logged as skipped.
Requests are paced at 1.5 seconds by default. On a 429 it backs off exponentially from 5 seconds up to a 30-second cap, honoring Retry-After headers, and pauses the run after repeated 429s from the same host.
GYF0311Trace academic claims to original publications and raw data.
SigmacodeatIngest external ideas and source materials into a connected knowledge graph with provenance tracking.
SigmacodeatSaves notes, files, and piped input into a searchable knowledge database.
tronghieuRun scheduled watchlist discovery and return a JSON summary of new paper candidates.