Generate Week Substack (Weekly Orchestrator — Substack pipeline)

/generate-week's sibling for Substack, covering both formats in one run: one Article (default target 1/week) and several Notes (default target 3/ week). Same shared Post-Ideas/ pool as the other two orchestrators. Never approves or schedules/publishes anything itself — only sequences the skills that do, ending at /publish-substack's manual handoff instead of a Buffer call.

Arguments

• week (optional) — passed through to both planning skills.
• article_count (optional) — passed to /plan-week-substack-article. Default 1.
• note_count (optional) — passed to /plan-week-substack-note. Default 3.

Process

1. Plan the week's Substack angles

1. /research-topic — reuse this week's existing research/idea pool if /generate-week or /generate-week-x already ran; otherwise run a scan sized for one deep article + a few lighter Notes (start at 6).
2. /generate-ideas (default range).
3. /plan-week-substack-article [week] — may honestly find no idea deep enough this week; that's a valid outcome, not a failure.
4. /plan-week-substack-note [week] [note_count].

If the Notes slots come up short, run one targeted /research-topic <pillar> 2 → /generate-ideas → /plan-week-substack-note retry, capped at 2 extra rounds. The Article slot does not get a forced retry loop — an empty article slot this week is reported honestly, not chased.

2. Generate the article, if one was assigned

One subagent (Agent tool, general-purpose, run_in_background: false) given the idea's full context (same shape as /generate-week's subagent briefing) and instructed to read and follow directly: .claude/skills/ write-draft-substack-article/SKILL.md, then .claude/skills/ critique-draft-substack-article/SKILL.md, then .claude/skills/ generate-visual-substack-article/SKILL.md. Same retry-on-duplicate instruction as /generate-week, capped at 2 retries. Append/update its content-index.md row (platform: substack-article).

3. Generate each Note, one subagent at a time, in day order

Same per-slot subagent pattern as /generate-week-x, pointed at .claude/skills/write-draft-substack-note/SKILL.md then .claude/skills/ critique-draft-substack-note/SKILL.md then .claude/skills/ generate-visual/SKILL.md (platform: substack-note). Same 2-retry cap. Append/update content-index.md rows (platform: substack-note).

4. Combined review, then manual-publish handoff

1. /review-drafts with no draft-id — shared across whatever orchestrators ran this session, run once total.
2. /publish-substack once, for whatever ended up status: approved with platform in {substack-article, substack-note}. This produces copy-ready Substack-Ready/ notes and asks the user, per item, whether it's actually been published — it does not assume publication happened.

After both, update content-index.md's status for each Substack row.

5. Final report

A table (article + Notes together): date | platform | topic | category | viral_score | visual format — plus a variety check and final counts (scheduled has no Buffer meaning here; report ready_to_publish vs. confirmed-published counts instead).

Hard rules

• Never skip /review-drafts before /publish-substack.
• Never call any Substack API — /publish-substack is manual by design.
• Never mark anything published without the user's explicit confirmation during /publish-substack.
• Never exceed 2 retries per slot; never pad the week's counts, for either the article or the Notes.