okhp3-linkedin-post

okhp3-linkedin-post

Portable, identity-agnostic Agent Skill.

Public-package data boundary

This package and its fixtures contain no execution account, person, project, campaign, topic, bias, opinion, or private URL. Use placeholders in public examples and obtain all such context from the current run only.

Drafting, from a chosen angle to finished output.

Scope

This package drafts a LinkedIn post only. It does not publish, schedule, manage comments, or create content for another platform.

Process

1. Input. A chosen angle (core insight + evidence + category, from okhp3-linkedin-angles' registry) or a directly-named topic. For repository activity, require the public ref, bounded observation window, timezone, retrieval time, and source locations before making a count or “since yesterday” claim.
2. Draft. Write the post. Lead with the angle's core insight. Do not use throat-clearing or a generic hook such as “Ever wondered...”. Keep a pending contest, award, or launch outcome pending; never write hope, preparation, or audience voting as a win.
3. CTA and links. Choose one primary destination. If the user supplied multiple public surfaces, include secondary links only when each has a distinct named role and the post remains scannable. Do not replace supplied URLs, invent destinations, or turn a comment-shaped request into a standalone campaign post.
4. Voice pass. Hand the draft to okhp3-linkedin-voice for the brand-rule pass (em dashes, paragraph consolidation, ROY density, link routing, uncertainty, and requested hashtag count).
5. Public-context scrub - final gate, hard, non-negotiable. Run references/public-context-scrub-checklist.md against the voice-polished draft. This runs LAST, after voice, because voice polish can itself introduce specifics. No draft leaves this skill without passing this gate, regardless of how confident earlier steps were.
6. Output. Return the finished draft, evidence note, primary CTA, secondary-link roles, and gate report. If anything was caught and removed at step 5, note only its category. State that publication remains a separate, explicitly authorized host action.

Order matters

Draft -> CTA/link check -> Voice -> public-context scrub -> output. Not scrub-then-voice: voice polish happens on content, then the final safety check happens on the actual thing that is about to be published, not an earlier draft of it.

Relationship to angles

If no angle was provided and the user just says "write a post about X" with no prior okhp3-linkedin-angles pass, that's fine — angles is for triage of EXISTING work, not a mandatory gate before every post. Direct topic requests skip straight to drafting.

Output contract

Return one paste-ready LinkedIn draft followed by a compact evidence and gate report. The draft must contain no employer-identifying material, no unsupported factual claims presented as facts, no settled language for an undecided outcome, and no unresolved public-context findings. Preserve any user-specified hashtag count exactly. If the requested claim cannot survive the scrub, return a generalized version or stop with the specific category of blocker.

Validation