Add httpwg chair-tooling skill (agenda subcommand)

[mnot]

Adds a repo-local Claude skill, httpwg, that captures how the chairs build
an httpbis IETF meeting agenda so it can be re-run each meeting (and shared
with co-chairs / future chairs).

The agenda subcommand walks the workflow:

• pin down session logistics (date, venue-local time, room, Meetecho links)
• pick the drafts in play and verify each one's real status
• weight time by activity, keeping each session within its own slot
• name only confirmed presenters
• push to gh-pages
• draft the call-for-presenters mail (never send)

It's built around the ietf-llm meeting_sessions / draft_status /
draft_authors tools, and treats the multi-session case (two slots, the day
block repeated) as first-class. It lives under .claude/ (dot-prefixed, so
Jekyll won't publish it).

The IETF 126 agenda itself is already on gh-pages; this PR is just the
skill.

---

AI-generated (Claude Code, Opus 4.8), high human supervision. It was
written by working a real task end-to-end with the chair — drafting the
IETF 126 agenda — and then capturing that workflow as a skill, iterated over
several rounds of chair feedback (format conventions taken from the most
recent agendas, multi-session layout, decoupling from any specific mail
tool, and wiring to the new ietf-llm lookups). The chair reviewed the
content before this PR.

[mnot]

Review — httpwg skill

Verdict: approve. Well-structured skill that follows the authoring conventions, and every checkable claim holds up against the repo.

Solid

• Description triggers well. States what it does, when, and gives concrete phrases ("start the IETF 126 agenda", "fix the times on the agenda"); dependencies (ietf-llm, mail) are declared. Good altitude — 301 lines, explains the why ("mostly verification, not authoring") rather than barking rules.
• Claims verified against reality. The skeleton matches ietf126/agenda.md and ietf125/agenda.md (heading, the tabbed time/room line, the Meetecho full/onsite pair, Administrivia boilerplate, * N min - [Title](url) - Presenter bullets). The "don't edit README, CI regenerates it" claim is real — lib/dir_listing.py runs from .github/workflows/dir_listing.yml, matching the "update indices" commits. gh-pages as published branch, the Co-Authored-By trailer, and the CODE.md pointer all check out.

Minor (non-blocking)

1. Section taxonomy is looser than the skeleton implies. The skeleton shows only Active Drafts + Other Topics, but ietf125 also carries a Related Work section and orders Other Topics before Active Drafts. The skill does say "follow the most recent agenda, not this skeleton," so it's covered in spirit — but a reader could treat the two-section, fixed-order layout as canonical. Naming Related Work as another legit section and noting the order isn't fixed would close the gap.
2. "Often two sessions" vs. recent reality. The multi-session handling is genuinely good and worth keeping first-class, but ietf125 and ietf126 both ran a single session, so "httpbis often gets two (and historically that's been the norm)" reads a touch strong against the last two meetings.

One thing to consider: scope this as chair tooling

The skill never says out loud that it's chair (or co-chair/delegate) tooling — it just assumes that context ("the chair's email", "when the chair approves"). I'd suggest guidance, not a hard gate: a coded "are you a chair" check is unenforceable, and the two consequential steps are already constrained by access — you can't push to gh-pages without commit rights, and you can't read confirmations or draft the mail without the chair's mailbox. A non-chair who invoked it gets at most a local, unpublishable draft. So one sentence near the top stating it presupposes chair authority is enough; no enforcement machinery needed.

Out of scope

ietf125/agenda.md:22 has a pre-existing typo — "Dict Hardt" → "Dick Hardt". Already committed, unrelated to this PR.

---

AI-generated (Claude Code, Opus 4.8); reviewed and directed by the chair, who requested the review and specified its scope. Factual/format claims were verified against the repo.

[mnot]

Thanks for the review. Addressed all three in 7d5b8e2:

1. Section taxonomy — added a paragraph noting the section set and order aren't fixed: Related Work and one-offs like AD-Requested Feedback are called out, and that recent agendas sometimes lead with Other Topics; the skill now says to take sections/order from the latest agenda, not the skeleton.
2. "Often two sessions" — softened to note recent meetings have been single-session and to check rather than assume, while keeping the multi-session handling first-class.
3. Chair tooling — added a sentence up front stating it presupposes chair authority, with your reasoning that the consequential steps are already access-gated, so no enforcement is needed.

Left the Dict Hardt typo in ietf125/agenda.md alone as out of scope; happy to fix it in a separate one-line PR if wanted.

AI-generated (Claude Code, Opus 4.8); the chair reviewed and directed these changes.