Prompt hub Half B: surface .claude/skills + the prompts submodule as prompt posts

[omars-lab]

Context

Half A of the Prompts hub shipped: /craft/software-development/prompts catalogs 17 kind: prompt posts (now on /initiatives) via <Catalog kind="prompt" />. Half B is surfacing the rest of the prompt corpus on the hub. Parked 2026-07-03 ("hold on creating more prompt posts for now") — capturing the scoping so it's not lost.

Intent (from the user)

1. Posts for the skills too — each .claude/skill becomes a post (same pattern as the prompt-logs).
2. The prompts submodule "naturally integrated as prompt posts in the right organization" — each prompt in the public repo github.com/omars-lab/prompts becomes a first-class prompt post, filed by area, not just a link out.

Candidate content (scoped)

• 54 .claude/skills (SKILL.md each) — this repo's internal SDLC toolchain (deploy-site, reorganize-content, manage-hubs, ...).
• ~35 prompts in the PUBLIC prompts submodule (github.com/omars-lab/prompts, licensed, already framed for sharing): analyze/ author/ bootstrap/ brainstorm/ draw/ heal/ meta/ organize/ plans/ refactor/ rules/ solve/.
• NOTE: .claude/commands and .claude/agents are empty (0 files) — the original task's "commands/subagents" framing is moot.

Open decisions (resolve before building)

1. Generated vs hand-authored — a re-runnable script emitting ~89 posts from source (title/area/description from frontmatter/README, stays in sync) vs hand-authored narrative per post vs generate-then-enrich. ~89 items is a lot to hand-write.
2. Organization — one Prompts hub grouped by area (logs + skills + submodule prompts intermixed by what they DO) vs a separate Skills hub (a new skill kind, paralleling how prompt/pattern/technique hubs work) for the internal tooling.
3. Exposure (the real blocker) — the skills include infra/secret-touching tooling (deploy-site, manage-infrastructure, manage-cloudflare-access) whose publication reveals how the private infra works. Decide: all-public vs curated non-infra subset vs plan-all-but-flag-the-risky-ones for explicit go/no-go.

Suggested first step when resumed

Resolve the three decisions above (an AskUserQuestion pass), then likely a generate-prompt-posts.js generator over .claude/skills/*/SKILL.md + prompts/<intent>/*.md, wired into generate-assets, filing each post by area onto the chosen hub(s).

Filed from the Prompt-hub Half-B backlog (task #32); references the durable-hub system (manage-hubs skill) and the kind: taxonomy (blog-kinds.json).