# Signal — LinkedIn Authority Engine > The LinkedIn vertical of ACOS. Draft, critique, repurpose, and schedule LinkedIn posts and articles in a specific creator's voice — for a personal brand and a company brand — then learn from what performed. Runs the Excellence Loop, drafts everything, and NEVER auto-posts (a human ships). Use for "/signal", "write a linkedin post", "linkedin article", "repurpose this for linkedin", "plan my linkedin week", "dream100", "become a linkedin top voice". - Skill: `frankxai-agentic-creator-os/signal-linkedin-authority-engine` (Agent Skill, multi-file: 7 files) - Install (CLI): `npx skillmds@latest add frankxai-agentic-creator-os/signal-linkedin-authority-engine` - Raw SKILL.md: https://api.skillmd.com/api/skills/frankxai-agentic-creator-os/signal-linkedin-authority-engine/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: frankxai (https://skillmd.com/u/frankxai-agentic-creator-os) - Updated: 2026-09-10 - Page: https://skillmd.com/skills/frankxai-agentic-creator-os/signal-linkedin-authority-engine --- # Signal — LinkedIn Authority Engine The productized LinkedIn layer of ACOS, and the flagship vertical of **The Studio**. Signal is a thin, sellable configuration of the general [Excellence Loop](../../docs/architecture/EXCELLENCE_LOOP.md) — the same eight steps, wearing a LinkedIn skin. Two things make Signal different from "an AI that writes LinkedIn posts": 1. **It runs a critique + learning loop.** Every draft passes an independent voice/claim/hook check before it's shippable, and every _shipped_ post's performance is written back so next week's drafts are grounded in what actually worked — not a blank prompt. 2. **It never posts for you.** LinkedIn's terms forbid automated posting through unofficial means. Signal drafts, critiques, and _stages_; a human clicks post. This is the compliant design and the simpler one. ## The two-brand model Signal always operates under one of two voices, loaded from [`personas.md`](personas.md): - **Personal** — the individual's authority voice (first-person, opinionated, career/insight-led). - **Company** — the brand's voice (we/product-led, proof and customer-outcome-led). Same engine, two lenses. Pass `brand: personal` or `brand: company`; default is `personal`. Never blend the two in one draft — pick the lens first. ## Modes (the command surface) Invoke via `/signal `. Each mode is a segment of the Excellence Loop. | Mode | Loop step | What it does | |---|---|---| | `plan` | Ideate | Reads the voice corpus + recent winners → a week of ranked angles mapped to pillars. | | `draft` | Craft → Execute | One idea/source → a LinkedIn draft in the chosen brand voice, staged as a draft. | | `repurpose` | Execute | Fetch a great external clip/post (Firecrawl) → a **credited** reframe. Highest-ROI motion. | | `critique` | Critique + Verify | Runs `integrity-guard` (brand/claim/schema) + a hook/format check. Returns PASS / warn / FAIL. | | `ship` | Interconnect | Exports the passed draft to a Typefully/LinkedIn **draft** via official API. **A human posts.** | | `learn` | Learn | Takes supplied post analytics → writes the graded outcome to the store → tunes next week. | | `research` | Ideate | Sweeps the source lanes (papers / videos / pdfs / posts) and vets them. See `research-sources.md`. | | `carousel` | Execute | A post → brand-locked carousel slides via multimodal + `canvas-design`. | | `infographic` | Execute | A post / data → a brand-locked infographic. | | `engage` | Distribute | Dream100 newest posts → substantive comment drafts. **A human posts** each. | | `index` | — | Index your existing content (posts / talks / repos) into the searchable library. | If no mode is given, route by intent: a pasted URL → `repurpose`; "plan my week" → `plan`; a raw idea → `draft`; "who should I engage" → `engage`. **Command surface:** `/linkedin ` (memorable front door) and `/signal ` are the same engine. ## The non-negotiables - **Never auto-post.** No mode calls a posting endpoint. `ship` produces a draft a human confirms. This is a hard stop inherited from the root `CLAUDE.md` (posting to LinkedIn is human-gated) and from LinkedIn ToS. - **Critique gates ship.** `ship` refuses a draft that hasn't passed `critique`. Fail-closed. - **Credit is mandatory on `repurpose`.** A reframe of someone else's clip names the source, on its own line, every time. - **No fabricated metrics.** Follower counts, view counts, "top voice" claims obey the Metrics-Truth rule — verify or omit. `learn` only writes numbers a human supplied from real analytics. ## LinkedIn format discipline (the format pass) Applied to every draft, from `knowledge.md`: - Plain text — LinkedIn renders no markdown. No `#`, `*`, or `-` bullets that won't render; use line breaks and unicode where needed. - Hook in the first 1–2 lines (before the "…see more" fold). The scroll stops here or nowhere. - One idea per 1–2 lines, generous line breaks — LinkedIn is skimmed on mobile. - ≤ 3 hashtags, at the end. One link, on its own line near the close (links suppress reach — place deliberately). - Match the persona's emoji/exclamation rules exactly (personal and company differ — see `personas.md`). ## The operating manual - [`DAILY_LOOP.md`](DAILY_LOOP.md) — the daily rhythm (≈45 min of your time: ideate → draft → design → critique → ship → engage → learn) + weekly cadence. - [`research-sources.md`](research-sources.md) — the best-of-best intake lanes (papers / videos / pdfs / posts), the vetting bar, and the signal→post pipeline. - [`agent-team.md`](agent-team.md) — the multi-agent org (lead + researcher / ideator / drafter / designer / engagement + independent critic). - [`../../docs/architecture/SIGNAL_ROADMAP.md`](../../docs/architecture/SIGNAL_ROADMAP.md) — the road to Top Voice + the product stages. ## The knowledge base (the moat) [`knowledge.md`](knowledge.md) holds the reusable intelligence Signal draws on: the hook taxonomy, the format playbooks (text post · document/carousel · article · repurpose), and the Dream100 method. This is the asset that compounds — it grows every time `learn` records what worked. ## What to build next (this pack is a scaffold) Per `L99_BLUEPRINT.md`, Signal is only _real_ once the loop closes. The build order: 1. A `content` table (state spine) so `draft`/`ship` persist instead of vanishing. 2. `ship` wired to the Typefully/LinkedIn official draft API. 3. `learn` wired to the Success Oracle store so performance tunes the next `plan`. Until then, Signal drafts and critiques honestly — and says so.