Run the content engine for a founder who is growing an audience to sell a product. Three jobs share the hook-retain-reward spine. The durable organic system decides what to post each week. Paid ad creative buys a cold stranger's attention with one sharp lever, and live viral research feeds a current trend into the system as fuel. A piece serves one audience through one lever, and a test reads one clean signal.

Posting and every engagement action (like, follow, reply, repost, schedule, DM) are external mutations that write to a live platform. This skill researches and drafts only: stop and get the user's explicit approval before any posting or engagement action. Draft, label the draft unposted, and leave the world unchanged.

Pick the reference for the job and read it before drafting:

references/content-system.md: the durable organic system, covering pillars, the hook-retain-reward post structure, format fit per platform, weekly cadence, repurposing, and the give-before-you-sell ratio.
references/ad-angles.md: paid ad creative, from the eight-lever angle taxonomy and 3-second hook patterns through format fit and message-market-match to the test and its failure modes.
references/viral-research.md: live viral research in the browser, spanning the manual research method, the viral-pattern taxonomy, adapt-not-copy, X-versus-Instagram norms, the post-and-test loop, and the ethics and ToS line.

Steps

Frame the audience, the offer, and the job. Write the founder's niche, the one target buyer this round serves, the product the audience is grown to sell, the felt pain or desire that buyer carries, and which job runs now (organic system, paid ad creative, viral research, or a combination). For paid creative, add the awareness stage (unaware, problem-aware, solution-aware, product-aware, most-aware). Write in the buyer's language, on the platform where that buyer already spends time; a global trend localizes to the buyer or it reaches no one who buys. This step is done when niche, buyer, product, pain-or-desire, and the named job are each written in one line.
Choose the lever per piece. For the organic system, name 3 to 5 content pillars sited where founder expertise crosses buyer pain on ground adjacent to the product. Paid creative takes 3 to 5 distinct angles from the eight-lever taxonomy (pain, desire, curiosity, social proof, authority, contrarian, us-vs-them, transformation). Viral research scans the live X, Instagram, and TikTok surfaces in the browser per the ethics line and captures 8 to 12 spiking posts, extracting each to its transferable pattern; a daily run records the captures and renders a ranked radar with scripts/trends.py report. This step is done when the job's unit exists: 3 to 5 pillars each with an audience payoff, or 3 to 5 angles each naming a different lever, or at least 3 captured posts each reduced to a one-line pattern stripped of its subject.
Write the hook and the body on the shared spine. Draft each piece as hook, retain, reward (organic and viral) or as a 3-second hook plus a beat-by-beat script or slide sequence (paid), using a named hook pattern. The opening line aims a tension at the viewer and promises only what the body will cash. Draft the whole piece the way the founder would say it out loud. This step is done when each piece carries a distinct hook line that would stop a target stranger on frame one with no sound and no brand name, plus a retain body that pays out at least one specific, non-obvious, screenshot-worthy beat, and an explicit reward or call-to-action.
Fit the format to the platform, then gate the fit with the script. Assign each piece one platform-native format from the format-fit tables: the standalone post and the thread on X, the carousel, the Reel, and the story on Instagram, plus the static image and the UGC creator-style format for paid creative. A format earns its place by reinforcing the lead lever. When a piece ships as a designed visual (carousel slides, covers), lock the founder's visual reference first and build on its extracted system (see the visual-identity section of the content-system reference), and deliver an editable artifact, a per-slide PDF or the source file, alongside whatever flattened images ship. Then collect the drafted set into a spec file (one entry per piece carrying its platform, caption, hashtags, and aspect) and run scripts/content-lint.py check <spec.json>; the script enforces the per-platform caption length, hashtag cap, and aspect ratio. This step is done when each piece names one platform and one format that reinforces its lead lever, and content-lint.py check exits zero on the spec.