- name
- update-blog-refs
- description
- Scan blog posts and suggest related cross-links. Use when the user invokes /update-blog-refs.
- disable-model-invocation
- true
Scan blog posts and suggest related links between them. Follow these steps exactly:
## Consequential advice
Before recommending a consequential cross-link, follow the `Advice gate` in
`dissenter`; routine relevance matches stay direct.
When the gate applies, first say that you are using `/dissenter` and why.
## Step 1: Read all posts
Use a blog directory supplied by the calling skill when present. Otherwise, read it from `docs/agents/blog-directory.md` in the current repo (or run `/setup-devinat1-skills` first).
Read every post except `index.md` and the `images/` directory. For each post, note its title (filename without `.md`) and core topic.
## Step 2: Identify related posts
For each post, identify which other posts are meaningfully related. Two posts are related when:
- They discuss the same tool, system, or concept
- One is a sequel, reference card, or deep dive of another
- They approach the same problem from different angles
- One describes building something the other uses
Posts are NOT related just because they share a broad theme like "AI" or "coding." The connection should be specific enough that a reader of one would genuinely benefit from reading the other.
When the calling skill supplies a focus post, consider every post but propose only links to or from the focus post. Otherwise, propose relevant changes across all posts.
## Step 3: Check existing related sections
For each post, check whether it already has a `## Related` section at the bottom. Note which posts already have one and what links they contain.
## Step 4: Present proposed changes
Show the user a summary of all proposed changes:
- Posts that need a new `## Related` section added
- Posts with an existing section that should have links added or removed
- Posts with no meaningful connections (these get skipped)
Format each proposal as:
```
**Post Title**
Add: [[Link 1]], [[Link 2]]
Remove: [[Link 3]] (if applicable)
```
Wait for user approval before making any changes.
## Step 5: Apply approved changes
For each approved post, add or update the `## Related` section at the very end of the file, preceded by a horizontal rule:
```
---
## Related
- [[Post Title]]
- [[Another Post Title]]
```
If the post already has a `## Related` section, update it in place rather than adding a duplicate.
## Conventions
- Heading is always `## Related`
- Each link is a bullet with an Obsidian-style `[[wikilink]]`
- No descriptions — just titles
- Bidirectionality is not enforced — each post's list is independent
- Posts with no meaningful connections omit the section entirely
## Automatic Jev check
When two already-read posts are a proposed cross-link pair, first identify the concrete connection normally, then send only their titles, relevant excerpts, and stable IDs after reading the existing `typesafe-ai` skill and its current API documentation. Only do this with operator authorization to disclose minimized evidence to TypeSafe; remove credentials and unrelated private data, and retain the ordinary workflow when consent or access is unavailable. Score reader benefit against five ordered descriptions (returned 0–4): broad shared theme; weak adjacent topic; useful but non-specific connection; specific supporting or contrasting connection; direct sequel, reference, deep dive, or dependency.
Use comparable scores only to rank already-proposed pairs, never as all-pairs discovery. On ambiguity, unavailability, or disagreement, retain the ordinary relevance judgment and approval gate; Jev never edits Markdown.
GitHubで見る