Social content

Create unpublished drafts from approved source material. Keep every non-obvious claim traceable to a real, named, linkable source; distinguish verified facts, approved positioning, user-provided opinion, inference, and unknowns; and leave publication decisions to the user.

Boundaries

• Do not invent metrics, outcomes, testimonials, partnerships, capabilities, pricing, urgency, or personal experience.
• Do not publish, schedule, post, reply, vote, contact users, access social sessions, or create accounts.
• Research current facts only when explicitly requested and approved. Cite named, linkable sources and preserve important caveats.
• Do not impersonate people or communities, disguise promotion, or use fake scarcity, outrage bait, undisclosed sponsorship, or instructions to evade platform or community rules.
• Do not turn a goal, forecast, plan, or aspiration into a reported result.
• Do not own LinkedIn-specific post drafting, carousel production, publishing, engagement, lead follow-up, or workspace state; hand those requests to the corresponding linkedin-* skill.
• Do not use unrelated private or internal communications as source material; this workflow is for user-provided or approved public-content drafts.
• Before presenting or saving a public draft, check supplied material for secrets, credentials, confidential business details, and personal data. Do not carry sensitive content into public copy; flag anything that needs the user's explicit publication decision.
• Do not hardcode personal or product identity. Use supplied brand rules and assets, and ask for the exact name, title, and domain before rendering a footer.
• Do not create files, start browsers, render images, or install tools until the user explicitly approves that work.

Workflow

Read references/workflow.md for the full content-pack, research, review, file-output, blog, and completion-checklist workflow. For visual assets, also read references/visuals.md.

1. Confirm the channel, audience, objective, topic, and source material. If a content-pack input is missing, ask for it instead of guessing.
2. If factual research is requested, build a short evidence ledger with source, claim, status, and caveat. If evidence is weak, label the draft opinion.
3. Draft for the selected channel using the compact guidance below. Load humanizer after the structure is complete when that companion skill is available; otherwise perform the same factual and voice-preservation audit directly.
4. Fact-check every claim, label the draft unpublished, and present it for review. Revise before saving.
5. After explicit approval, save under <project-root>/sm_<slug>/posts.md with channel headings and a Sources section. Resolve <project-root> before writing; never publish or commit.

Channel targets

Channel	Shape
X	Concise post/question; 3-6 hashtags; URL counts as 23 characters.
Reddit	Relevant context, useful contribution, transparent affiliation.
Hacker News	Factual/technical; lead with what was made and why.
Product Hunt	Maker context, solution, limits; no manufactured proof.
Blog/newsletter	Evidence, examples, takeaway, sources.

Use the source's own words only for necessary exact names, titles, or numbers; paraphrase everything else. Lead with the strongest real finding, explain the mechanism, and end with a concrete implication rather than a generic CTA.

Blog and visual branches

For a blog from existing social content, expand the core insight into a thesis, evidence-backed sections, implications, and sources. Save blog.md only after approval. For visual assets, read references/visuals.md before proposing deliverables. When a connected Canva capability is available, also read references/canva-workflow.md and use Canva as the preferred authoring path. Wait for approval before creating or exporting any visual.

Every handoff includes sources, claims needing approval, required disclosures, and wording that must not be presented as fact.