linkedin-ghostwriting

Applies when ghostwriting a named executive's next LinkedIn posts. Produces three ranked, typed options per exec, each a few lines, each with an attachment, each cleared by an anti-repeat gate.

This skill exists to fix the two failure modes of exec ghostwriting: posts that repeat what the exec already said, and posts that read like marketing copy instead of the person. It runs in one pass and hands the exec ready-to-post options to approve by hand. It never posts anything.

Required inputs

Gather these before drafting. A thin input set produces generic posts.

• A voice card per exec - their real diction, the arguments they own, what they will never say, and one or two of their best past posts. Build it from their actual posting history, not a guess. Format and starter questions are in references/voice-card-and-cadence.md.
• A post ledger - every post that has gone out, with hook, core argument, attachment, and any performance note. This is what the anti-repeat gate reads.
• A fresh signal feed - this week's news, competitor moves, earnings, and regulatory shifts on the exec's beat, plus repost candidates. Verify each peg's date at the primary source, not the commentary's date.

The play

1. Read the history first. Before generating anything, read the ledger and the last several live posts: what argument, what attachment, what landed. This feeds the anti-repeat gate and keeps cadence honest.
2. Mine the signal, then decide the angles. Read the whole signal feed and generate the angles from it. Treat any pre-suggested angle as one raw input, never the plan.
3. Run the anti-repeat gate. Diff every candidate angle against the ledger and the exec's off-limits list. Kill or re-angle anything on the off-limits list, in the same argument family as a recent post, or inside the same-argument cooldown window. A rejected or deleted past option counts as a rejection - never re-serve it. The full gate logic, cooldown windows, and cadence targets are in references/anti-repeat-and-cadence.md.
4. Lock the three types and the attachments. Assign one of each type per exec: a news share plus comment, a repost, and a thought-leadership piece with a graphic. For the graphic, recommend single image versus carousel and hand over a ready-to-paste image-generation prompt. The format gates, attachment rules, and image-prompt spec are in references/format-and-attachments.md.
5. Check messenger and lane fit. The exec must plausibly follow that source and their audience must care about it - not merely "is the argument sound". Route the peg to the exec whose lane it is (regulatory to the policy voice, product to the builder, numbers to the finance voice). Two execs never comment on the same article.
6. Draft in voice, then two finishing passes. Draft each option against the voice card and closest samples. Then strip AI tells to a clean baseline, and add one or two real human moves so the post reads as the person, not a clean bot. One or two moves, never invented detail.
7. Score, rank, and deliver. Score each on hook strength and on commercial value (audience precision, proof, narrative fit, why-now). Below bar gets one revision. Deliver three ranked options per exec with attachments and, for the graphic, the image recommendation. Dependencies and posting-order notes travel with the delivery, never on the post itself.
8. Close the loop. After the exec picks, edits, or posts, append the chosen post to the ledger - it re-enters play only after the cooldown window, never sooner. If they rewrote it, that rewrite becomes the new gold line - update the voice card. Any angle the exec killed or rejected goes to the off-limits map as a permanent block. (Ledger entries return after the cooldown; off-limits entries never do.)