LinkedIn Brand Direction

Runs on: local-writes - reads your audit and the algorithm reference, then writes brand-direction.json outside the repo and refreshes the local pack-state pointer.

This is the personalisation engine of the LinkedIn pack. It does not give generic advice. It reads the composition of your actual network, your goal, and the dated algorithm facts, and returns a concrete direction tied to evidence from your own data. linkedin-post then writes posts from that direction.

Known limitation: Algorithm facts are reverse-engineered and dated. If skills/linkedin-pack-references/linkedin-algorithm.md's last_verified stamp is over ~90 days old, re-verify before trusting specifics.

Inputs (in priority order)

1. Network composition. Prefer audit.json from linkedin-power-audit (richest: role distribution, content lanes, brand signals). If it is absent, fall back to network-scan.json from linkedin-network-scan (carries the qualified-lead composition, ICP, and totals). If neither exists, stop and tell the user to run linkedin-power-audit (best) or at least linkedin-network-scan first - do not invent a composition.
2. The goal. Ask one question if it is not already known from core/identity.md or the prior session: "What is this brand for - winning clients, getting hired, or being known in a space?" The direction aims at that goal.
3. The algorithm reference. Read skills/linkedin-pack-references/linkedin-algorithm.md. If it is missing, say "algorithm reference not found, using conservative defaults" and use only format-and-cadence basics.

The output contract (fixed schema - do not invent field names)

Write brand-direction.json into the user's scan/audit output folder (the same folder the scan wrote to, OUTSIDE any git repo). The schema is fixed so linkedin-post can read it without guessing:

{
  "goal": "clients | job | brand",
  "topic_lane": "the lane the network already rewards, with the evidence (e.g. 'ops and supply-chain - 41% of your qualified network')",
  "positioning_angle": "one line - the angle that lane gives you",
  "format_mix": [
    {"format": "document carousel", "why": "tops the reach ladder; your lane is explainer-friendly"},
    {"format": "native short video", "why": "second on the ladder; good for a face-to-camera take"}
  ],
  "cadence": "e.g. 3x/week",
  "first_three_posts": [
    {"hook": "first line that earns the 'see more' tap", "angle": "what it argues", "format": "document carousel"},
    {"hook": "...", "angle": "...", "format": "text"},
    {"hook": "...", "angle": "...", "format": "native short video"}
  ],
  "evidence": [
    "topic_lane: 41% of the 220 qualified connections hold an ops or supply-chain title (network-scan.json)",
    "format_mix: document carousels lead the reach ladder (linkedin-algorithm.md, data-backed)"
  ]
}

The machine-readable contract is brand-direction.schema.json. Do not add stable fields without updating that schema and the tracked acceptance test.

Rules for filling the schema

• topic_lane must name the lane the data supports and carry the number behind it. Derive it from the role/industry concentration in the composition, not from a guess about the person.
• positioning_angle is one line, grounded in the lane plus the goal.
• format_mix follows the current evidence: documents/carousels are strong and underused, images are reliable, strong text can work, and video is not assumed to rank second. Each entry carries a one-line why.
• cadence comes from the algorithm reference's conservative current range of 2 to 4 posts per week. State a specific number.
• first_three_posts are three concrete starts, each a hook plus angle plus a format from the mix. The hooks must pass the "see more" test (would a reader tap?).
• evidence is the spine. Every claim in the direction traces to either a real number in the composition file or a labelled fact in the algorithm reference. If you cannot trace a claim, cut it. No invented percentages, no made-up audience.

No hallucination

The whole value here is that the direction is real. Do not invent a network composition, a percentage, or an audience the data does not show. If the composition is thin (a small or unscored network), say so and give a narrower, honest direction rather than padding it with generic advice. Carry the export's limits forward: title-only, point-in-time, no firmographics, email mostly blank.

Handoff

Write or refresh ~/.founder-os/linkedin-pack-state.json with latest_output and brand_direction pointing to the completed bundle. Tell the user the direction is written to brand-direction.json and that linkedin-post will read the pointer automatically on the next post request. If they want to act now, route straight to linkedin-post.

Files

• skills/linkedin-pack-references/linkedin-algorithm.md - the dated algorithm facts this skill reads.
• brand-direction.schema.json - exact output contract.
• linkedin-power-audit - produces the audit.json this skill prefers as input.
• linkedin-network-scan - produces the network-scan.json fallback composition.
• linkedin-post - the consumer of brand-direction.json.