Pixi Vault to Pixi Wiki Publishing Model¶

pixi-vault is the private source repository. pixi-wiki is the public generated mirror. The boundary is intentional: authoring truth and public output are separate artifacts.

Source side¶

pixi-vault/
├── Knowledge/              # reusable concept authoring
├── Projects/               # project/application source truth
├── Wiki Compiler Maps/     # routing contracts
└── wikis/<slug>/           # compiled namespace source

The wikis/<slug>/ layer is compiled/curated source inside the vault. It is reviewable in Git and Obsidian, but agents should edit Knowledge/, Projects/, or Wiki Compiler Maps first unless intentionally patching compiled output.

Public side¶

pixi-wiki/
├── llms.txt
├── llms-full.txt
├── index.json
├── search.html
├── updates.html
├── updates.json
├── sitemap.xml
├── site.css
├── raw/<slug>/...
└── wiki/<slug>/...

llms.txt is the compact agent registry. llms-full.txt is the full concatenated corpus. index.json is the machine-readable registry. raw/<slug>/ preserves Markdown, wiki/<slug>/ exposes human-readable HTML, search.html/updates.html provide human retrieval and freshness surfaces, and the local MCP server exposes read/search tools over the same KB files.

Human pages render Markdown with breadcrumbs, visible frontmatter metadata, raw Markdown links, report-a-mistake links, namespace sidebars, heading anchors, table of contents, and previous/next navigation. The goal is a two-surface publication model: humans browse/search the web wiki and update history, while agents use raw Markdown, llms.txt, index.json, and local MCP tools over the same KBs.

Publication workflow¶

1. Update canonical source in Knowledge/, Projects/, or Wiki Compiler Maps.
2. Compile/curate into wikis/<slug>/.
3. Run namespace linting against pixi-vault/wikis.
4. Run the pixi-wiki generator.
5. Run public-output tests.
6. Push both repos as needed.
7. Verify live GitHub Pages URLs with HTTP 200 and expected content tokens.
8. Run browser-level verification for pages affected by CSS, layout, navigation, or JavaScript. HTTP checks alone miss computed-style and runtime failures.
9. For local MCP changes, run the MCP self-test and a real stdio client smoke test.

Clean rebuild policy¶

Old root flat pages in pixi-wiki are removed from the canonical public contract. The clean mirror should keep root-level global surfaces such as the homepage, registries, search, Updates, sitemap, shared assets, and docs, plus /raw/<slug>/... and /wiki/<slug>/... namespace trees.

No old concept-*.html, projects-*.html, knowledge.html, projects.html, maps-of-content.html, root.html, agent/, or legacy/ surface should reappear without a deliberate compatibility policy and regression coverage.

Cross-namespace relationships¶

• Agent Workflows documents crew workflows affected by source-truth boundaries, Knowledge Pack Routing, and markdown-first agent memory.
• Eval Trace evaluates whether agents overfit stale context or skip live verification.
• Local AI Infrastructure will matter when RAG or local models consume compiled wiki pages.

Source¶

Compiled from Projects/Hermes Mission Control/PRD - Pixi Vault Namespace Compiler.md and Wiki Compiler Maps/Namespace Wiki Compiler Map.md.