| name | linkedin-brand-direction |
| description | Turn your real LinkedIn network plus your goal into a personalised, algorithm-aware content direction. Trigger on "what should I post on LinkedIn", "what's my content lane", "give me a LinkedIn content direction", "what does my network reward", or after a power audit when the user wants a brand strategy. Reads the network composition from linkedin-power-audit's audit.json (or falls back to the network-scan.json composition), the person's stated goal, and the shared algorithm reference, then writes a defined-schema brand-direction.json that linkedin-post consumes. Every claim traces to a real fact in the user's own network - no generic tips, no invented numbers.
|
| why | Generic LinkedIn advice ignores the one thing that matters - what THIS person's network actually rewards. Conditioning the direction on real network composition plus the dated algorithm facts turns 'post more carousels' into 'your network is 41% ops leaders, so own the ops-systems lane in document carousels three times a week'. |
| enhance | Run linkedin-power-audit first so the direction reads from a full audit.json (network composition, content lanes, brand strengths) instead of the lighter scan composition. The richer the input, the more specific the lane evidence. |
| summary | A content direction conditioned on your real network, not generic tips. |
| allowed-tools | ["Read","Write","Bash"] |
| mcp_requirements | [] |
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)
- 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.
- 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.
- 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')"
