generate-content

name	generate-content
description	Generate platform-native content for all channels from a kept idea. Produces TikTok EN, TikTok DE, Instagram, and LinkedIn drafts with voice profiles applied and critic review completed. Use when user says '/generate-content', 'create content for idea X', or 'generate posts from this topic'.
argument-hint	[idea-id]
allowed-tools	["Read","Write","Bash","Glob","Grep","Agent","SendMessage","Skill"]

generate-content — Full Pipeline Content Generation

One command turns a kept idea into platform-native drafts across 4 channels, with critic review applied before Robin sees anything.

Usage

/generate-content <idea_id>

<idea_id> — UUID of a kept idea from data/content.db.

What It Does

1. Read the kept idea from data/content.db (must have status = 'kept').
2. Research pass (D-06) — produces a structured brief at data/research/<idea_id>.md via NotebookLM. Cached on re-runs. Sets research_thin flag if sparse.
3. Pick slide track — branded or personal, from idea.visual_approach or inferred from the idea title.
4. Repo screenshot pre-flight — if the idea references a GitHub repo, the repo-screenshot subagent captures a clean PNG of the repo card for use as an image-overlay slide.
5. Per platform: spawn writer → run critic loop → render slides → save draft.
   • TikTok EN, TikTok DE: writer produces a slides_spec for the chosen track. Rendered via generate-branded-slides or generate-personal-slides.
   • Instagram: writer produces a caption only. Slides reuse TikTok EN PNGs (cropped if personal track).
   • LinkedIn: always branded. Writer produces a post text + branded slides_spec. Rendered via generate-branded-slides.
6. Save drafts to data/content.db with status = 'critic_approved'.
7. Report the summary table to Robin.

Implementation

Spawn the content-orchestrator agent:

Agent(content-orchestrator):
  "Generate content for idea <idea_id>."

The orchestrator handles everything: research pass, track selection, repo-screenshot dispatch, writer/critic loops, slide rendering via the two new skills, and DB writes. You do NOT run any scripts directly here.

Present the orchestrator's summary output to Robin when it completes.

Output to Robin

Content generation complete for idea: <idea_id> — "<title>"
Track: <branded|personal>   Repo screenshot: <yes|no>

Platform      | Slides                | Critic            | Draft ID
------------- | --------------------- | ----------------- | --------
TikTok EN     | personal N slides     | critic_approved   | <uuid>
TikTok DE     | personal N slides     | critic_approved   | <uuid>
Instagram     | reuses TikTok EN      | critic_approved   | <uuid>
LinkedIn      | branded N slides      | critic_approved   | <uuid>

Files: media/output/<idea_id>/

Next: /approve to review and schedule.

If drafts hit max-iters: **Warning:** X draft(s) did not pass critic — best-scored iteration kept with did_not_pass_critic=true.

Flags Robin Will See

• research-thin (amber): research returned insufficient info; writer leaned on idea fields. Advisory.
• did-not-pass (red): critic loop hit its 2-revision cap without scoring ≥ 8 on every dimension. Best iteration kept.

Neither blocks approval.

Key Rules

• ALWAYS read storytelling-framework.md and the voice profile before generating (writer enforces this).
• Slide skill is deterministic by platform. No track-picking.
  • LinkedIn → generate-branded-slides (11x cream/clay templates, 1080×1350). Always. Never personal.
  • TikTok EN/DE → generate-personal-slides (real photos from media/images/tiktok/catalog.json, 1080×1920). Always. Never branded templates on TikTok.
  • Instagram → generate-personal-slides (cropped from TikTok EN spec, 1080×1350). Always. Never branded.
• TikTok DE reuses EN photos — same photo_keywords per slide, only text and caption change.
• Instagram reuses TikTok EN spec — re-rendered at 1080×1350. Only its own caption is fresh.
• Critic runs as a separate agent invocation per platform (D-12) — fresh context.
• No post publishes without Robin's approval. This skill only generates and gates drafts.

Dependencies

Agents

• .claude/agents/content-orchestrator.md — coordinator
• .claude/agents/writer.md — per-platform writer
• .claude/agents/critic.md — scoring and revision
• .claude/agents/repo-screenshot.md — captures GitHub repo PNGs for image-overlay slides

Skills

• .claude/skills/generate-branded-slides/ — universal branded carousel renderer (1080×1350)
• .claude/skills/generate-personal-slides/ — photo + text overlay renderer (TikTok 1080×1920, IG 1080×1350)
• .claude/skills/writing/data/voice-casual.xml — TikTok/Instagram voice profile
• .claude/skills/writing/data/voice-linkedin.xml — LinkedIn voice profile
• .claude/skills/writing/data/storytelling-framework.md — narrative structure

Utility Scripts

• scripts/db-read-idea.js — read idea from DB
• scripts/db-write-draft.js — save draft to DB
• scripts/research/run-notebooklm.sh — research pass
• scripts/generate-tiktok-slides.js — node-canvas compositor (used by personal slides)
• scripts/screenshot-github.py — Playwright fallback for repo-screenshot subagent

Data

• data/content.db — idea backlog and draft storage
• data/research/<idea_id>.md — research brief cache
• media/output/<idea_id>/ — generated slides per platform
• media/images/tiktok/catalog.json — personal photo library (used by personal slides)
• media/overlays/github/ — repo-screenshot cache