LinkedIn Signal Analyser

Public setup guide: vibha-ramprakash.github.io/Linkedin-Manager

Example dashboard: vibha-ramprakash.github.io/Linkedin-Manager/signal.html

The hosted dashboard is an example workspace only. Live LinkedIn research is deliberately local: clone the repository, follow SETUP-GUIDE.md, or paste INSTALL-PROMPT.md into a local coding agent.

A filmable research workspace with two explicitly separate modes:

• Example run works offline with locally stored example records.
• Live run uses the locally authenticated Agent Reach LinkedIn route and a visible, read-only browser session.

Live mode does not grant unlimited access, bypass platform controls, connect with people, or send messages.

Open the example workspace

Open signal.html in a modern browser. No server or LinkedIn account is required. The top bar remains labelled Example data · no live LinkedIn session.

Start the live workspace

First complete the one-time LinkedIn sign-in:

uvx mcp-server-linkedin@latest --login

Then start the local bridge from this folder:

node live-server.mjs

Open:

http://127.0.0.1:4317

Choose Live run. The workspace performs a harmless profile read before showing LinkedIn connected · live read. A configured MCP alone is not treated as proof of a working session.

Side-by-side recording

1. Complete login before the take.
2. Start live-server.mjs.
3. Choose Live run so the visible MCP browser opens.
4. Place Signal Analyser on the left and the MCP browser on the right.
5. Enable Recording privacy if the dashboard identities should be masked.
6. Start screen capture and click Run live research once.

The browser visibly navigates LinkedIn with a 350 ms action delay. The dashboard receives progressive events as profiles, companies, posts, fit scores, themes, and drafts become available. No additional click is required unless LinkedIn presents a login confirmation, CAPTCHA, or security checkpoint.

Blur real names and profile photos in the LinkedIn half before publishing the recording.

For the next take, choose 3, 5, or 10 from the Next run control beside the People Finder action. Run 10 fresh requests up to ten people. When a live run already contains records, Add 10 to run retains the current evidence, excludes usernames already present, and recomputes the qualified cohort and trends from the merged evidence set.

The requested amount is a maximum, not a guaranteed result count. The installed LinkedIn people-search tool does not expose pagination or a result-limit parameter, so LinkedIn may return fewer inspectable profiles for a query. The completion state reports the returned count against the requested count instead of presenting under-delivery as a dashboard error.

Live-run boundaries

• The saved batch size controls both fresh and additive runs and can request 3, 5, or 10 profiles.
• Profiles are processed serially.
• Tools are restricted to people search, person profiles, company profiles/posts, post search, and session closure.
• The server binds only to 127.0.0.1.
• Browser cookies and session data are never sent to the page.
• Sending, connecting, inbox access, and publishing are not available.
• If a live search returns zero results, the interface stays empty. Example records are never mixed into a live run.
• Run live search starts fresh; Add N to run preserves and deduplicates the existing v2 run.
• Earlier sourced profiles are retained in the unified 48-hour working set with Previous run provenance. They remain usable in People Finder, Fit Checker, Evidence Writer, and Review Queue, but do not enter Trend Scanner aggregates without normalized evidence.

Workspace views

Reel beat	Product view
Research team and outputs	Research Overview
Agent Reach, MCP, and context	Setup & Context
Small sourced batch	People Finder
Recent activity and themes	Trend Scanner
ICP and timing checks	Fit Checker
Drafts and content ideas	Evidence Writer
Human approval	Review Queue

Unified 48-hour lead workspace

Live mode deduplicates the current v2 run and the preserved 19 September profile receipts into one 15-lead working set. People Finder shows all 15 with Current run or Previous run provenance. Fit Checker ranks the same set by fit score, then timing, and exposes All, ICP matches, and Outside ICP filters. Matches are auto-approved for drafting; exceptions are isolated for human review. Public promotions, new roles, company moves, expansions, and launches appear as sourced timing events and never increase ICP fit. Evidence Writer exposes all 15 as selectable draft tabs, and Review Queue resolves their retained source receipts.

Trend Scanner lists all 15 under Lead coverage, but only profiles with normalized post evidence can influence the evidence map. A legacy profile with profile/company receipts remains visible as Profile sources only; it is not treated as dated trend evidence.

Package contents

• signal.html: offline workspace plus the live-mode client
• index.html: interactive GitHub Pages setup guide and dashboard entry point
• INSTALL-PROMPT.md: copy-paste Agent Reach and LinkedIn setup prompt
• live-server.mjs: localhost-only LinkedIn bridge and event stream
• trend-engine.mjs: evidence normalization, cohort qualification, date bucketing, deduplication, and theme metrics
• trend-engine.test.mjs: deterministic evidence and additive-run fixtures
• content-map-engine.mjs: evidence-aware pillar, format, goal, hook, readiness, and post-build recommendations
• content-map-engine.test.mjs: deterministic content-planning and claim-safety checks
• config/mcporter.json: visible, slowed, keep-alive LinkedIn MCP configuration
• SETUP-GUIDE.md: install, authentication, and implementation guide
• THIRD-PARTY-NOTICES.md: attribution for adapted content-planning concepts
• QA-REPORT.md: functional, credibility, and accessibility checks
• skills/people-finder/SKILL.md
• skills/trend-scanner/SKILL.md
• skills/fit-checker/SKILL.md
• skills/evidence-writer/SKILL.md

Data handling

Example-mode edits and review decisions are stored in browser local storage. Evidence-first live results use the linkedin-signal-analyser-live-v2 key. Previous profile records are copied into linkedin-signal-analyser-prior-runs-v1 for the separate reference panel; legacy trend aggregates are never imported into Trend Scanner. Source receipts include the permitted URL and retrieval time returned by the connector. No LinkedIn credentials are written into this package.

Evidence-first Trend Scanner

Trend Scanner uses only profiles that meet the supported role, company/market, and fit requirements. A market-post search may enrich that qualified cohort but cannot replace one; zero qualified profiles produce zero themes. It deduplicates source receipts, excludes undated items from 7/14/30-day totals, and recomputes appended runs from the merged evidence set instead of adding old totals.

Four keyboard-operable views make the research easier to inspect:

• What people discuss maps recurring questions, complaints, launches, and changes by source breadth.
• Recent activity shows the dated person posts, company posts, and market-search receipts in chronological order.
• Product launches shows only posts with explicit launch, release, rollout, or availability language and labels them as observed announcements—not purchase intent.
• Industry shifts groups wider market and company-change signals while stating that a current-window sample is not a velocity claim.

The coverage map uses unique sources horizontally, represented people and companies vertically, and deduplicated mentions for bubble size. Selecting a bubble exposes observation metrics, a cautious interpretation, an editable content angle, and direct source links. The hosted example includes clearly labelled illustrative receipts; live mode uses the same views with LinkedIn source URLs and retrieval times returned by the local connector.

Evidence-led Content Map

Evidence Writer turns each sourced theme into a usable post plan. It ranks text, carousel, poll, short-video, and single-image fit; assigns a pillar and engagement goal; proposes a non-repeating hook pattern; and shows the post structure, CTA, missing proof, sample risk, and evidence coverage. A concentrated or single-source signal is visibly downgraded instead of presented as a ready market insight.

The planning vocabulary and hook-pattern discipline are adapted from sergebulaev/linkedin-skills. This package does not execute that repository or send content to LinkedIn; see THIRD-PARTY-NOTICES.md.

Evidence Writer now has two product tabs: Outreach drafts and Content map. Content Map builds a six-slot Monday/Wednesday/Friday schedule with topic, format, pillar, goal, readiness, source count, and production state. Text posts, PDF carousel slide copy, static-image treatments, and short-video scripts use format-specific generation actions. Generated states and the selected asset persist locally; every output remains a draft requiring human review.