Wiki Consume/Produce Split

The wiki organizes content into two top-level concerns:

• consume/ — External input: YouTube transcripts, article extracts, concepts and entities from consumed content. Everything Hermes digests from outside sources.
• produce/ — Output: blogs synced from Joseph's Mac, expense tracking, and what Joseph and Hermes create together (books, products, blogs, self, library, plans).

Architecture

wiki/
├── index.md
├── log.md
├── schema.md
├── readme.md
├── agents.md
│
├── consume/                   ← External input (Hermes writes here)
│   ├── concepts/              Extracted concepts from consumed content
│   ├── entities/              People/orgs from external sources
│   ├── highlights/            Key highlights from sources
│   ├── meta/                  Meta-tools & references about knowledge management (permanent, not extracted)
│   ├── raw/                   Staging: raw source files (ephemeral, deleted after processing)
│   ├── tracking/              Ingestion metadata
│   └── transcripts/           Raw YouTube transcripts
│
└── produce/                   ← Output (blogs synced, Hermes writes hermes/)
    ├── blogs/
    │   ├── jjude/             ← SYNCED FROM MAC (jjude.com). Hermes NEVER writes here.
    │   └── ctofieldnotes/     ← SYNCED FROM MAC. Hermes NEVER writes here.
    │
    └── hermes/                ← What Joseph and Hermes create together
        ├── books/             (hs, always-be-shipping, word-in-world, etc.)
        ├── blogs/
        │   ├── ctofieldnotes/ (ideas, style-guide)
        │   ├── drafts/        (unified blog draft dir)
        │   ├── social-media/  (cross-cutting social pipeline for both blogs)
        │   └── prompts/
        ├── commercial/        (coaching, [client], [client], vc)
        ├── self/              (assessments, decisions, frameworks, goals, retros)
        ├── library/           (writing-concepts, writing-themes)
        ├── products/          (manufacturing-saas)
        └── plans/

The Consume Pipeline

Content flows through consume/ in a standard pattern:

1. Raw source arrives — YouTube transcript, PDF, article URL

2. Extract outputs — concepts/, highlights/, entities/ (one file per source)

3. Discard raw — the raw transcript/PDF is deleted after extraction

The sources: frontmatter in extracted files should reference the source URL, not a local file path. This allows raw files to be deleted without creating dead references.

YouTube:  transcript → extract → delete transcript
Article:  PDF/URL → extract → delete raw
Paper:    PDF → extract → delete raw

Why delete raw? Joseph never re-reads raw transcripts or PDFs. The extracted highlights/concepts/entities are the permanent artifact. Keeping raw files adds clutter without value. Exception: consume/meta/ — Meta-tools and references about knowledge management itself (specs, standards, tool evaluations). These are NOT processed through the extract→delete pipeline. They are permanent reference files that Joseph bookmarks for future evaluation. Content here follows Joseph's adoption philosophy: capture the reference, analyze relevance, but adopt only if it delivers 10x benefit — otherwise it's just tool-hopping. See references/consume-meta-template.md for the entry format.

The Mac Sync Boundary

produce/blogs/ is synced FROM Joseph's Mac. Hermes reads these directories but NEVER writes to them. Anything Hermes writes there will be overwritten on next sync.

Hermes-extracted insights from blog posts live in produce/hermes/library/ — specifically:

• writing-concepts/ — 227 concepts extracted from Joseph's blog posts
• writing-themes/ — 7 cross-cutting themes discovered across his writing

Write Permissions

Location	Who writes	Synced?
consume/	Hermes	❌ No
produce/blogs/	Joseph's Mac	✅ Yes
wealth/expense/	Hermes	❌ No
produce/hermes/	Joseph + Hermes	❌ No

When This Skill Activates

Use this skill when:

• Setting up wiki structure
• Understanding where to write content
• Deciding where extracted concepts/themes from blog posts should live
• Troubleshooting broken wikilinks after directory moves
• Reorganizing produce/hermes/ directory structure

For setting up the sync infrastructure itself (rclone, cloud storage, cron jobs), see the cloud-sync skill.

Pitfalls

• Never write to produce/blogs/ — it will be overwritten by Mac sync. See references/mac-sync-boundary.md for the full split pattern.
• Extracted blog insights go to produce/hermes/library/, not to consume/ (they're derived from output, not external input)
• Wikilinks across consume/produce use [[../consume/concepts/name]] from produce, or [[../produce/hermes/books/hs]] from consume
• Root-owned files can't be mv'd or rm'd by Hermes — use cp -a + manual cleanup by the user. See references/mac-sync-boundary.md.
• After reorganizing hermes/ directories, scan all wiki docs for stale path references — wiki/agents.md, wiki/readme.md, and internal READMEs often contain absolute or relative paths that break. Use search_files with the old path pattern across the entire wiki before declaring the move done.
• Joseph dislikes underscore-prefixed folders (_meta, _archive, etc.) — use descriptive names like library/ instead.
• Per-project Gumroad materials live inside the book directory (books/hs/gumroad/), not in a top-level gumroad/ folder.
• consume/meta/ is for reference only — when bookmarking tools, specs, or standards for future evaluation. Don't process these through the extract→delete pipeline. Include a mapping table showing what's relevant to our wiki and what's not, plus Joseph's assessment of adoption timing.
• Joseph's adoption threshold: 10x benefit — he captures references but only adopts tools/standards that deliver transformative value. Otherwise it's tool-hopping that prevents achieving anything. When writing consume/meta/ entries, note the potential value but respect that adoption is a separate decision.