Private Source Corpora as Executable Skills

·https://github.com/badass-courses/second-brain
repoaiknowledge-managementsecond-brainparavaultpi-notesagent-skillsretrievaltypescript

maps the joelclaw Vault + pi-notes pattern to a private-source/public-skill boundary for agent-usable knowledge

Joel dropped this from Slack as “a bunch of PARA/Second Brain stuff for shitrat to use,” which is exactly the interesting part: badass-courses/second-brain is not just notes. It is a private, source-grounded operating layer for turning Forte Labs / Tiago Forte material into something agents can query, classify, and act on.

The repo draws a useful boundary: paid/source material stays private, while distilled executable behavior can live in badass-courses/skills. That keeps the corpus from leaking into portable public/project skills while still letting Pi, pi-notes, and the Obsidian Vault use the ideas operationally. No bullshit copy-paste knowledge laundering.

The clever bit is that it treats PARA as a runnable classification and review workflow, not a folder naming religion. Given a .brain note, the vertical slice classifies it as Project / Area / Resource / Archive, explains the call with source-backed evidence, suggests missing fields, generates a next action or review cadence, and links back to chunks/full articles.

This is useful for joelclaw because it is the same shape as the wider system: source material, chunked retrieval, .brain documents, an Effect CLI, a SQLite database, and agent-facing skills. The public-safe artifact is the pattern: keep source truth close, expose tiny executable interfaces, and make every agent recommendation trace back to receipts.

Key Ideas

  • badass-courses/second-brain is a private paid-source corpus, .brain wiki, retrieval database, Effect CLI, Pi extension, and operational skill bundle for PARA / Second Brain execution.
  • The repo separates source truth from portable execution: full paid/source material stays in the private corpus, while distilled skills belong in badass-courses/skills.
  • The current corpus is generated from a Roam Research V1 source set and includes three entries, 232 chunks, a nomic-embed-text embedding model, and a vacuumed .brain/data/forte.db around 39 MB.
  • The V1 workflow turns a .brain note into an operational decision: classify it, explain why, suggest missing fields, propose a next action or cadence, and link back to evidence chunks.
  • The source policy is the important guardrail: preserve paid/source text privately, keep visible claims source-grounded, label pi-notes adaptations as adaptations, and avoid committing transient SQLite WAL/SHM files.
  • The repo uses Bun commands for status, source listing, Roam Research ingest, search, chunk context, note classification, database vacuuming, and workflow inspection.

Links

← All discoveries