X Bookmarks to Wiki Pipeline
Kevin's X bookmarks do not die as saved links. They flow into the wiki as compiled, cross-referenced pages, the capture loop made literal.
Routing
Open this page when explaining or refining the X bookmark ingestion loop. Route operational changes to Capture Ingest Protocol, Source Compile Automation Setup, and x-bookmark-absorb; keep this story page focused on why the pipeline exists and what good output looks like.
Pipeline Map
| Stage | Owner | Output |
|---|---|---|
| Sync | ft sync / source compile |
Fresh X bookmark export under raw/x-bookmarks/ |
| Enrich | source-compile + x-bookmark-absorb |
Tweet, replies, linked artifacts, media, and author context |
| Resolve | Filing Decision Tree + Capture Ingest Protocol | Correct target pages and cross-links |
| Compile | Agent writer | Durable wiki pages with citations, tweet embeds, and source artifacts |
| Index | build-index, qmd update, qmd embed |
Searchable graph state for future brain-first lookup |
Source Tweet Artifacts
The pipeline has a defined path. Per AGENTS.md, X bookmarks sync via fieldtheory (ft sync), copy into raw/x-bookmarks/, and then get ingested into the wiki. Source: AGENTS.md The ingest is not a dump. Capture Ingest Protocol requires the full graph integration: the right pages, cross-links, citations, backlinks, index, and search update, not a single stub. Source: Capture Ingest Protocol
X Bookmarks: Dedalus (Oct 2025 – May 2026) is the output shape: bookmarked tweets from the Dedalus orbit compiled into a themed page with takeaways, dated and attributed by handle, organized into fundraising, product, and culture sections rather than a flat list. Source: X Bookmarks: Dedalus (Oct 2025 – May 2026) On the read and write side, X-Agent splits cost-free reads (cookie-auth CLI via Agent Reach) from official-API writes (xmcp), so the agent can search and pull bookmarks for free and only spend credits when posting. Source: X-Agent
The reason it is a pipeline and not a habit is No One-Off Work: if Kevin has to be asked twice, it should already be a skill on a schedule, so capture is codified rather than repeated. Source: No One-Off Work The payoff is The Brain-Agent Loop: signal gets compiled once and stays searchable, so lookup is brain-first before any external call. Bookmarks become durable knowledge instead of a backlog nobody revisits.
The codified form is deep absorption, not a hub entry. A dedicated x-bookmark-absorb skill exists to push bookmarks past a single list into dedicated pages, cross-references, resolver updates, and index entries, so a saved tweet becomes a node in the graph rather than a line in a file. Source: AGENTS.md
As of 2026-06-30, the sub-daily owner is source-compile: local Codex syncs X bookmarks every four hours, runs enrichment, scans the author's self-thread and artifact replies, resolves any repo/docs/demo links, researches promoted entities online, and then updates hub pages, dedicated wiki pages, skills, tool routers, _index.md, qmd, state.json, and wiki/log.md. The key shift is that a bookmark is treated as a research object: tweet + replies + linked article/repo/media + independent sources. Source: automations/source-compile.md, 2026-06-19; User request, 2026-06-30
Nick Spisak (@NickSpisak_) described the same workflow in a viral thread: a twitter-bookmarks Claude Code skill that reads Chrome-profile bookmarks, prioritizes by recency, reviews against brand voice and mission, drafts a comprehensive implementation plan as atomic todos, gets human approval, then implements. A follow-up in the thread generalizes the method: start Claude Code or Codex in plan mode, give it a repository, ask it to research the codebase, interview it about benefits and features, ask for an atomic implementation plan, conduct human review, iterate until the plan is right, then approve implementation. Kevin's kevin-wiki pipeline extends that with enrichment (scripts/enrich-x-bookmarks.ts), RESOLVER filing, media review, source review, and mandatory cross-links. Source: X/@NickSpisak_ thread, 2026-03-16 A follow-up bookmark framed ETL patterns for topic research using qmd, agent-browser, and Obsidian — the same stack this repo uses. Source: X/@NickSpisak_ thread, 2026-04-05
Timeline
- 2026-04-07 | X-Agent created: two-backend read/write split for X automation. Source: X-Agent
- 2026-05-11 | Capture ingest protocol codified for the paste-and-sync workflow. Source: Capture Ingest Protocol
- 2026-05-17 | Dedalus X bookmarks compiled into a themed wiki page. Source: X Bookmarks: Dedalus (Oct 2025 – May 2026)
- 2026-05-31 | x-bookmark-absorb skill noted as the codified deep-absorption step beyond hub entries. Source: AGENTS.md
- 2026-05-31 | Story page compiled. Source: compiled from wiki graph
- 2026-06-12 | Top-25 absorption linked @NickSpisak_ twitter-bookmarks skill recipe and ETL-pattern follow-up to this pipeline story. Source: x-bookmark-absorb, 2026-06-12
- 2026-06-19 | Daily
source-compileautomation upgraded to pull X bookmarks, enrich linked artifacts and author replies, research promoted entities online, and write results back into the wiki/skill/router graph. Source: User request, 2026-06-19; automations/source-compile.md - 2026-07-03 | Deep-reviewed the @NickSpisak_ thread and preserved the exact plan-mode workflow: bookmark skill reads source, agent researches, outputs atomic todo plan, human reviews, then implementation is explicitly approved. Source: X/@NickSpisak_, 2026-03-16
- 2026-07-03 | Source-reviewed the @NickSpisak_ follow-up bookmark. The resolved X Article shell was login-gated and not useful, but the self-thread preserved the transferable ETL pattern: use summarization for YouTube transcripts, agent-browser for web collection,
qmdfor semantic search, and Obsidian/wiki analysis for pattern synthesis. Source: X/@NickSpisak_, 2026-04-05