メニュー
Audit cortex articles in 10-cortex/ for decay, missing required fields, broken Sources prefixes, and short TL;DR sections. Read-only by default. Invoke with /cortex-lint to surface issues without mutating any file.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | cortex-lint |
| description | Audit cortex articles in 10-cortex/ for decay, missing required fields, broken Sources prefixes, and short TL;DR sections. Read-only by default. Invoke with /cortex-lint to surface issues without mutating any file. |
| user_invocable | true |
| argument-hint | topic to lint (e.g., ai, career) -- omit to lint all topics |
A health check across 10-cortex/ that flags articles that have decayed, are missing required frontmatter, or violate the cortex article standard. Read-only -- it diagnoses, never edits. To fix issues, hand the report to /cortex-compile or edit the article directly.
Read 00-agentic-OS/paths.json for vaultRoot. All paths below are relative to it.
With topic argument: lint articles under 10-cortex/<topic>/.
Without topic argument: lint every article under 10-cortex/ except _index.md.
If 10-cortex/ is empty or missing, report: "No cortex articles to lint." and exit.
For each markdown file under the lint scope (excluding _index.md files):
If frontmatter parse fails, record the article as a parse-error finding and skip the rest of its checks.
Run each rule against every article. Each rule produces zero or more findings.
verified_at over 90 days oldverified_at is missing: record missing-verified-at.verified_at parses to a date older than 90 days from today: record stale-verified-at with the age in days.confidence: low: record low-confidence (informational; not a failure on its own, but surfaces articles that should either be promoted or removed).Required: title, description, topic, created, last_compiled, verified_at, confidence, staleness_signal, sources.
For each missing field, record missing-field:<field-name>.
## TL;DR heading: record missing-tldr.tldr-out-of-range with the word count. (Target is 50-100, with a buffer.)## Sources heading: record missing-sources.empty-sources.supports::, contradicts::, extends::, mentions::, inspired-by::): record untyped-source with the item text.## Connections heading: record missing-connections.[[...]] wikilinks: record empty-connections.staleness_signal is free-form text. Run a best-effort match: if the signal mentions a model name, API version, or product that the LLM knows has changed since verified_at, record staleness-signal-triggered with the matched phrase.
This rule is intentionally fuzzy. The LLM uses its own world knowledge to flag plausible decay; the user makes the final call.
Group findings by article. Within each article, order findings by severity:
| Severity | Findings |
|---|---|
| error | parse-error, missing-tldr, missing-sources, empty-sources, missing-field:title, missing-field:topic |
| warn | missing-field:* (for any other field), untyped-source, tldr-out-of-range, missing-connections, empty-connections |
| info | low-confidence, stale-verified-at, staleness-signal-triggered, missing-verified-at |
Articles with at least one error sort first. Within a severity bucket, articles with more findings sort first.
Write the report to stdout. Do NOT write a file -- the report is for the human to read in the chat.
cortex-lint report (YYYY-MM-DD)
Scope: <topic or "all topics">
Articles linted: N
Articles with findings: M
ERRORS (block re-compile until fixed)
10-cortex/ai/transformers.md
- missing-tldr
- missing-field:topic
10-cortex/career/grade-7.md
- parse-error: unexpected token at line 4
WARNINGS (fix when you next touch the article)
10-cortex/ai/rag.md
- untyped-source: "[[40-raw/youtube/foo.md]] -- pipeline overview"
- missing-connections
INFO (surface for review, not blocking)
10-cortex/ai/llm-routing.md
- stale-verified-at: 137 days old
- staleness-signal-triggered: signal mentions "GPT-4 turbo" which has been deprecated
Summary: 2 errors, 3 warnings, 2 info
Next: hand the error list to /cortex-compile to refresh those articles, or edit them directly.
If there are zero findings, report:
cortex-lint report (YYYY-MM-DD)
Scope: <topic or "all topics">
Articles linted: N
Articles with findings: 0
All clean. Nothing to fix.
/cortex-compile job.verified_at for you. That's a manual decision -- bumping the date is a claim that you re-checked the source.40-raw/. It only inspects 10-cortex/ articles.The ideaverse-OS build interview and capture orchestrator. Use when the user runs /ideaverse-os build to compile their identity into the lean 6 files, or /ideaverse-os capture to ingest a source. Harness-agnostic - works in any LLM CLI that can read project files.
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.
Use this skill to generate well-branded interfaces and assets for ideaverse-os, either for production or throwaway prototypes/mocks/etc. Contains essential design guidelines, colors, type, fonts, assets, and UI kit components for prototyping.
Use this skill to generate well-branded interfaces and assets for ideaverse-os, either for production or throwaway prototypes/mocks/etc. Contains essential design guidelines, colors, type, fonts, assets, and UI kit components for prototyping.
Conversational ingestion router. Accepts a YouTube URL, plain HTTP URL, a filename in 40-raw/web-clips/, or a free-text "I learned X" note. Routes to the right handler, refreshes 10-cortex/ via cortex-compile + cortex-connect, and logs to today's daily note. Invoke with /ideaverse-capture <url-or-text>.
Scan 10-cortex/ articles for unlinked mentions of other article titles and add [[wikilinks]]. Deepens the knowledge graph without needing an LLM for matching -- pure text search. Invoke with /cortex-connect [topic] after cortex-compile, or run standalone.