Social Kit

Social Kit

A thin orchestrator over three sub-skills — create-x-content (X variants), create-linkedin-content (LinkedIn variants), and graphics-studio (the graphic) — with generate-voice-guide as a recommended one-time prerequisite. The bundled init_run.py scaffolds the run folder and prints the orchestration plan; everything else is delegation to the sub-skills (the agent drafts/authors).

When to use

• "Give me X + LinkedIn posts and a graphic for [brief]."
• When you want copy and a visual together, in the user's voice.

How to run

1. Scaffold the run + get the plan:

python3 ${SKILL_DIR}/scripts/init_run.py --brief "Shipped a 10x faster lead scraper" \
  --output ${WORKSPACE}/content
# optional: --skip-x / --skip-linkedin / --skip-graphic, --format poster, --style ...,
#           --variants-x N, --variants-linkedin N

Creates content/YYYY-MM-DD-<topic-slug>/ (+ a graphic/ subfolder) and prints {topic_slug, folder, steps, voice_guide_preflight}.

2. Voice-guide pre-flight — for each non-skipped platform, ensure an X and a LinkedIn voice guide exist; for any missing one, delegate to generate-voice-guide, paste a path, or skip that platform. Never proceed without resolving it — generic posts are the failure mode.
3. Draft X variants (unless --skip-x) via create-x-content, writing into the run folder. Capture the most substantive variant for the graphic brief.
4. Draft LinkedIn variants (unless --skip-linkedin) via create-linkedin-content.
5. Pick the graphic format — use --format if given, else recommend 1–2 from the brief shape (poster / carousel / infographic / tweet / story / slides / chart) and ask.
6. Pick a style — --style if given, else the house default.
7. Generate the graphic via graphics-studio from a distilled brief (headline + 2–4 key beats + must-appear strings, drawn from the strongest X variant), output to graphic/.
8. Deliver a combined, cross-linked summary; post files as Agent Teams / Slack attachments.

Invoke each sub-skill's scripts by path, e.g.:

python3 ${SKILL_DIR}/../create-x-content/scripts/lint_post.py --file <variant> ...
node ${SKILL_DIR}/../graphics-studio/scripts/render.mjs --html <slide> --out <png> --format <fmt>

python3 ${SKILL_DIR}/scripts/init_run.py --help lists all flags.

Outputs

A single run folder with X variant files, LinkedIn variant files, and a graphic/ subfolder (HTML source + PNG exports + preview), plus a summary linking drafts to the graphic.

Credentials / env

• Required: none. The agent drafts the copy and authors the graphic HTML; all sub-skill render/lint scripts are keyless — the whole kit runs key-free by default.
• Optional (all paid, all with a fallback; belong to a sub-skill's path; skip flags drop any arm):
  • APIFY_API_TOKEN / PHANTOMBUSTER_API_KEY — voice-guide corpus scraping (generate-voice-guide). If set → auto-scrape; if not → paste-text fallback (the default).
  • UNSPLASH_ACCESS_KEY and generative-art keys (LEONARDO/STABILITY/FAL) — graphic image sourcing (graphics-studio). If set → real/generated imagery; if not → keyless picsum/ASCII (the default).

Notes & edge cases

• Never proceed without resolving the voice guide for each non-skipped platform.
• Distill the graphic brief from the strongest X variant (headline + 2–4 beats + verbatim must-keep strings like tool names, commands, prices).
• Pair drafts with graphics intentionally — a long mechanism variant suits a carousel; a short hype variant suits a poster.
• First-time setup is generate-voice-guide once; iterate on voice before iterating on copy.