Hugo site: AGENTS.md compliance

For Markdown in the jeffbaileyblog Hugo site (<repo>/hugo/), the rules in AGENTS.md are mandatory. Read the files every session; they change, and a remembered summary goes stale. This skill adds the mechanics AGENTS.md leaves implicit and three scripts so the checks are one command each.

Scripts (paths relative to this skill, ${CLAUDE_SKILL_DIR}/scripts/): banscan.py, fmcheck.py, verify-build.sh. They are the only copy: write-validate-article calls them by sibling path (../hugo-agents-compliance/scripts/), so keep their flags, labels, and exit codes stable, and rename or move them only together with that skill. Preflight: command -v hugo python3. If one is missing, print the install command (brew install hugo / brew install python) and report "tool missing", not "check failed".

1. Scope the task first

Decide which mode you are in, because it controls how much you touch:

• New post: apply every rule below, including SEO wording from seo-front-matter.md.
• Existing post (a build fix, a link fix, or "make it compliant"): compliance means the mechanical rules. Fix what was asked; quote or unquote fields as required (keep the wording); add required fields that are missing (lastmod = today, keywords, cover.alt, a category); and fix banscan BLOCK hits (emdash, banned phrase, inline link, bare ref) with the smallest rewrite of that sentence. Never change date:. Do not reword an existing title, description, category choice, or untouched prose. That is an editorial or SEO decision the user did not ask for, and many legacy posts predate the current rules. Put those ideas under "suggested, not applied" in your report so the user can opt in.

Profanity is the author's voice in either mode. A compliance pass never removes or softens it.

2. Read the rules in their real order