cortex-compile

name	cortex-compile
description	Compile source notes from 40-raw/ into structured cortex articles in 10-cortex/. Creates or refreshes LLM-maintained articles. Invoke with /cortex-compile [topic] after any ingestion event, or run standalone to process all pending sources.
user_invocable	true
argument-hint	topic tag to compile (e.g., ai, career, productivity) -- omit to process all pending sources

cortex-compile

The synthesis step that turns raw captures in 40-raw/ into lasting knowledge articles in 10-cortex/.

Path resolution

Read 00-agentic-OS/paths.json for vaultRoot. All paths below are relative to it.

Sources

Source folder	Content type
40-raw/youtube/	YouTube research notes written by yt-light-research
40-raw/web-clips/	Processed web clip notes written by web-clip-report
40-raw/harvest/	Knowledge extracted manually from domain folders
40-raw/plain/	Plain-text synthesis notes from ideaverse-capture

---

Step 1: Identify pending sources

With topic argument: filter sources to files whose topic: frontmatter matches the argument.

Without topic argument: collect all sources in the above folders that are NOT already logged as compiled in 40-raw/_index.md.

Read 40-raw/_index.md (ingestion log) to skip already-compiled sources.

Scale guidance: If there are 50+ source files, prioritize: files with status: pending first, then the 15 most-recently-modified, then any files matching known gaps in 10-cortex/. Do not read all files at once.

If no pending sources are found, report: "No pending sources in 40-raw/. Nothing to compile."

---

Step 2: Determine target articles

From each source file's topic: frontmatter, determine:

• Target topic folder: 10-cortex/<topic>/
• Candidate article: check if an article covering the source's main concept already exists in that folder.

For each source:

• If an existing article covers it at 70%+: plan a merge (update that article).
• Otherwise: plan a new article.

---

Step 3: Write or refresh articles

For each planned article, create or update the file at 10-cortex/<topic>/<slug>.md:

---
title: "[Concept Name]"
description: "[One sentence: what this is. One sentence: why it matters to the vault owner.]"
topic: "[topic-slug]"
sources:
  - "[source-path]"
created: YYYY-MM-DD
last_compiled: YYYY-MM-DD
verified_at: YYYY-MM-DD
confidence: low | medium | high
staleness_signal: "[one-line condition that, if true, means this article is stale]"
---

# [Concept Name]

One-sentence definition of what this is.

## TL;DR

50-100 words. What this is and why the reader cares. Enough to know whether to keep reading.
This section is required -- cortex-lint flags articles missing it.

## Summary

2-3 paragraphs. Written for the vault owner 6 months from now who needs a fast refresher.
Emphasize the "why" and key insights, not definitions. Ground it in their active work where possible.

## Key Facts

- Concrete finding or principle
- Concrete finding or principle

## Connections

- **Related:** [[other-cortex-article]], [[another-article]]
- **Used in:** (domain contexts where this applies)
- **Contrasts with:** [[contrasting-concept]] -- (one-line distinction)

## Sources

Use typed-edge prefixes so the relationship between source and claim is explicit:

- supports:: [[40-raw/youtube/light-research-slug.md]] -- one-line note on what this source supports
- contradicts:: [[40-raw/web-clips/clip-slug.md]] -- one-line note on what claim this source disputes
- extends:: [[40-raw/papers/paper-slug.md]] -- one-line note on what this source builds on
- mentions:: [[40-raw/youtube/passing-mention.md]] -- one-line note on a passing reference
- inspired-by:: [[40-raw/plain/seed-thought.md]] -- one-line note on the seed idea

Quality rules

• One concept per article. If an article is over 600 words, consider splitting it.
• Prefer updating over creating: new source 70%+ covered by existing article -> update, don't create.
• Every article must have at least one entry in Sources, using one of the 5 typed-edge prefixes.
• Every article must link to at least one other article in Connections.
• Every article must have a 50-100 word ## TL;DR section. cortex-lint enforces this.
• Write for the vault owner. If 00-agentic-OS/me.md is in context, use it to ground the "Used in" and TL;DR relevance.
• Confidence: high for primary research or official docs, medium for credible analysis, low for speculation or single source.
• verified_at: set to today's date when the article is created or substantively re-confirmed against current sources. cortex-lint flags articles where verified_at is older than 90 days.
• staleness_signal: a one-line condition that, if true, means the article is no longer accurate. Examples: "model name changes", "API version bumps to v2", "company is acquired or shuts down". cortex-lint surfaces these for periodic review.
• Do not invent. Only write facts supported by the source material.

---

Step 4: Update 10-cortex/_index.md

Create 10-cortex/_index.md if it doesn't exist. Update it to list all articles:

# Cortex Index

Last compiled: YYYY-MM-DD
Total articles: N

## <topic>

| Article | Description | Last compiled |
|---|---|---|
| [[10-cortex/<topic>/<slug>.md]] | <description from frontmatter> | YYYY-MM-DD |

---

Step 5: Update ingestion log

For each source that was compiled, add a row to 40-raw/_index.md:

| YYYY-MM-DD | <source-path> | <cortex-article-path> | compiled |

Create 40-raw/_index.md with a table header if it doesn't exist.

---

Step 6: Archive compiled raw sources

After logging, move each compiled source from 40-raw/<type>/filename.md to 99-archive/<type>/filename.md. Create the destination subfolder if it does not exist.

Update any cortex article that references [[40-raw/<type>/filename]] in its Sources section to point to [[99-archive/<type>/filename]].

Do NOT archive sources tagged keep-in-raw: true in their frontmatter.

---

Step 7: Report

cortex-compile complete:
- Topics compiled: [comma-separated list]
- Articles created: N
- Articles updated: N
- Sources archived: N
- Ingestion log: updated 40-raw/_index.md

List any article that was created for the first time so the user knows to run /cortex-connect to add wikilinks.