Linkify CodeGrid skill mentions

CodeGrid articles in the AI-agents series mention Claude Code skills by name (/commits, /pr-complete, …). This skill turns the first prose mention of each published skill into a link to the author's public collection, claude-resources, pinned to the latest commit hash.

Pinning to a hash (not main) is deliberate: the repo changes often, and a hash-based ref keeps the article's link pointing at the version that existed at writing time. The repo is introduced once in an earlier article (a [column] titled 「筆者のスキルコレクション」 in 9.md), so later articles just inline-link — no need to re-introduce the repo.

Workflow

Step 1 — Identify the target article

Use the article the user is editing (from conversation context) or the path they pass. It's a file like data/{series-slug}/{n}.md in the CodeGrid articles repo.

Step 2 — Run the scan

python3 $HOME/.claude/skills/dev-linkify-cc-resources/scripts/scan.py <article.md>

The script resolves the claude-resources repo ($HOME/repos/*/claude-resources), fetches and reads the latest origin/main hash, lists the published skills, and reports a LINK THESE block: the first prose mention of each skill that exists in the repo, with the ready-to-paste markdown. It already excludes mentions inside code blocks, mentions already linked, and names that aren't real published skills.

Pass --repo <path> if the repo lives somewhere other than $HOME/repos/*/claude-resources.

Step 3 — Apply the links

For each entry in LINK THESE, wrap the inline-code skill name in the suggested link, leaving the surrounding prose untouched:

…9回目の連載で紹介した`/commits`という…
→ …9回目の連載で紹介した[`/commits`](https://github.com/Takazudo/claude-resources/blob/<hash>/skills/commits/)という…

Edit the live article, not the scan output. After editing, the URL format is https://github.com/Takazudo/claude-resources/blob/<hash>/skills/<name>/ — keep /blob/ and the trailing slash (GitHub redirects directory /blob/ to /tree/).

What gets linked, and what doesn't

The scan's repo membership check enforces the editorial rule automatically:

• Linked: skills present in claude-resources/skills/, at their first prose mention.

• Not linked: names that are hypothetical examples in the prose (e.g. /fix-typo,

  /brighten-photo) or personal skills the author hasn't published (e.g. /price-research) — these simply aren't in the repo, so the scan won't surface them.

• Skipped: mentions inside fenced code blocks, and any skill already linked elsewhere in

  the article (only the first prose mention is linked, for readability).

If the user disagrees with a specific call (e.g. wants a second mention linked too, or wants to skip re-linking a skill already linked in an earlier article), follow their preference — the scan is a starting point, not a hard rule.