feat: import candidate brief from LinkedIn profile (#48)

README.md

@@ -57,6 +57,8 @@ PUPILA shells out to whichever AI CLI you already have authenticated locally —
 
 Drag a `.pdf` / `.docx` / `.md` / `.txt` CV onto the drop zone (or click **Choose file**). The CV is parsed locally and sent to your LLM CLI to generate a short candidate brief — who you are, what stack you work in, what kind of role you want, and what to avoid. The original CV stays on disk at `config/cv.<ext>` (gitignored) so **AI Apply** can re-attach it later.
 
+**No recent CV?** Click **Import from LinkedIn instead** (optional) and upload a LinkedIn profile PDF — on your LinkedIn profile, go to **More → Save to PDF**, then drop the downloaded file here. It runs through the same parse → LLM pipeline with a LinkedIn-tuned prompt, so the resulting brief is comparable to a CV-sourced one. (No login or scraping — the "Save to PDF" export is fully self-serve.)
+
 ### 3. Confirm the generated brief
 
 <p align="left">
@@ -84,7 +86,7 @@ Contributing rules and project invariants live in [`CONTRIBUTING.md`](./CONTRIBU
 > **Looking for today's matches?** → [`JOBS.md`](./JOBS.md) (auto-generated by the local aggregator).
 > Raw data lives in [`data/jobs.json`](./data/jobs.json).
 > Local RSS feed at [`data/feed.xml`](./data/feed.xml) (point your reader at the file:// path).
-> Prefer a UI? → `pnpm run ui` opens a local-only Vite dashboard at `http://127.0.0.1:5173`. Four tabs: **Jobs** (filter, search, sortable columns, click-to-expand rows over `data/jobs.json`), **Jinder** (Tinder-style swipe deck — right-swipe to queue an AI Apply, left-swipe to skip), **Profile** (drop a PDF/DOCX/MD CV to set up or refresh your candidate brief), and **Settings** (scheduler lifecycle, scoring profile regenerate, disk usage, apply queue). See [AI Apply](#ai-apply-per-job-optional) and [AI per-job review](#ai-per-job-review) below.
+> Prefer a UI? → `pnpm run ui` opens a local-only Vite dashboard at `http://127.0.0.1:5173`. Four tabs: **Jobs** (filter, search, sortable columns, click-to-expand rows over `data/jobs.json`), **Jinder** (Tinder-style swipe deck — right-swipe to queue an AI Apply, left-swipe to skip), **Profile** (drop a PDF/DOCX/MD CV — or import a LinkedIn profile PDF — to set up or refresh your candidate brief), and **Settings** (scheduler lifecycle, scoring profile regenerate, disk usage, apply queue). See [AI Apply](#ai-apply-per-job-optional) and [AI per-job review](#ai-per-job-review) below.
 
 ---
 
@@ -113,16 +115,19 @@ The friendliest path is the **first-run onboarding wizard**: run `pnpm run ui` o
 For the CLI-only path, run setup-brief directly:
 
 ```bash
-pnpm run setup-brief --file ~/Documents/cv.pdf      # PDF
-pnpm run setup-brief --file ~/Documents/cv.docx     # Word document
-pnpm run setup-brief --file ~/Documents/cv.md       # markdown
-cat resume.txt | pnpm run setup-brief               # stdin
+pnpm run setup-brief --file ~/Documents/cv.pdf          # PDF
+pnpm run setup-brief --file ~/Documents/cv.docx         # Word document
+pnpm run setup-brief --file ~/Documents/cv.md           # markdown
+pnpm run setup-brief --linkedin ~/Downloads/profile.pdf # LinkedIn "Save to PDF" export
+cat resume.txt | pnpm run setup-brief                   # stdin
 ```
 
+`--linkedin` runs the same pipeline as `--file` but tells the LLM the input is a LinkedIn profile export (so it ignores LinkedIn's boilerplate). Don't have a recent CV? On your LinkedIn profile, go to **More → Save to PDF** and pass the downloaded file. (A `--file` whose name contains "linkedin" is auto-detected as a LinkedIn source.)
+
 Or open the UI and use the Profile tab:
 
 ```bash
-pnpm run ui   # http://127.0.0.1:5173 → Profile tab → drop your CV
+pnpm run ui   # http://127.0.0.1:5173 → Profile tab → drop your CV (or "From LinkedIn")
 ```
 
 The auto-detected provider order is `claude` → `codex` → `gemini` → `opencode` (whichever is on `PATH` first). Override with `PUPILA_LLM=codex pnpm run setup-brief ...`. No API keys; uses your existing CLI subscription.

src/lib/brief-prompt.ts

@@ -0,0 +1,53 @@
+// Single source of truth for the candidate-brief summarization prompt.
+//
+// Both entry points that turn a CV-like document into config/candidate-brief.md
+// share this builder so they can't drift:
+//   - the `pnpm run setup-brief` CLI (src/setup-brief.ts)
+//   - the UI's POST /api/cv middleware (ui/plugins/brief.ts)
+//
+// The output contract is identical regardless of source — three short
+// paragraphs of plain markdown — so a LinkedIn-sourced brief is comparable in
+// quality to a CV-sourced one. Only the *framing* changes: a LinkedIn "Save to
+// PDF" export has a predictable structure (and predictable boilerplate) that
+// the LLM does better with when we name it explicitly.
+
+export type BriefSource = 'cv' | 'linkedin';
+
+// Shared three-paragraph contract + closing instructions. Kept verbatim across
+// sources so the resulting brief has the same shape no matter where the raw
+// text came from.
+const OUTPUT_CONTRACT = `Output ONLY three short paragraphs as plain markdown text. No preamble, no markdown fences, no headings, no commentary.
+
+PARAGRAPH 1 — Who they are: role, years of experience, primary location, primary stack/skills. Be concrete (frameworks, languages, tools they ship with regularly).
+PARAGRAPH 2 — What they're looking for: target seniority (senior / lead / staff / principal IC), domains/sectors of interest (web3, AI, fintech, etc.), location preference (remote-worldwide / remote-EMEA / hybrid in <city> / open to relocation).
+PARAGRAPH 3 — What to avoid: roles that look like a fit on paper but aren't. Examples: wrong specialty (backend if frontend, etc.), wrong level (junior, intern, exec), on-site only, US-only positions, support/solutions/devrel/GTM titles.
+
+Aim for 6-10 lines total. Drop anything that doesn't help a job-matching tool decide. Don't editorialize.`;
+
+const CV_INTRO = `You are summarizing the following CV into a short candidate brief that will be sent to an LLM each time the candidate's job-matching tool evaluates a posting. The brief decides whether the LLM agrees with the rule-based fit score.`;
+
+const LINKEDIN_INTRO = `You are summarizing the candidate's LinkedIn profile into a short candidate brief that will be sent to an LLM each time the candidate's job-matching tool evaluates a posting. The brief decides whether the LLM agrees with the rule-based fit score.`;
+
+// LinkedIn "Save to PDF" exports interleave profile content with structural
+// boilerplate (Contact section, "Page N of M" footers, endorsement/skill
+// counts, "Top Skills", repeated headers). Naming the source lets the LLM
+// ignore that noise and read the Experience/Education sections as a résumé.
+const LINKEDIN_PREAMBLE = `The text below was extracted from a LinkedIn profile exported via "Save to PDF". Treat it as the candidate's résumé. Ignore LinkedIn boilerplate — contact details, "Page N of M" footers, skill-endorsement counts, "Top Skills" / "Contact" section labels, and repeated page headers. Infer the candidate's current role, seniority, and stack from the Experience and Education sections.`;
+
+/**
+ * Build the summarization prompt for a CV or LinkedIn profile export.
+ *
+ * @param text     parsed plain text of the document (CV or LinkedIn PDF)
+ * @param source   'cv' (default) or 'linkedin' — only changes the framing
+ * @param maxChars hard cap on how much of `text` we forward to the LLM
+ */
+export function buildBriefPrompt(text: string, source: BriefSource, maxChars: number): string {
+  const label = source === 'linkedin' ? 'LINKEDIN PROFILE' : 'CV';
+  const intro = source === 'linkedin' ? `${LINKEDIN_INTRO}\n\n${LINKEDIN_PREAMBLE}` : CV_INTRO;
+  return `${intro}
+
+${OUTPUT_CONTRACT}
+
+${label}:
+${text.slice(0, maxChars)}`;
+}

src/setup-brief.ts

@@ -5,17 +5,24 @@
 // keeping it sharp directly improves the per-job verdicts.
 //
 // CLI:
-//   pnpm run setup-brief --file path/to/cv.pdf       # parse PDF (via pdfjs-dist)
-//   pnpm run setup-brief --file path/to/cv.docx      # parse DOCX (via mammoth)
-//   pnpm run setup-brief --file path/to/cv.md        # plain markdown
-//   pnpm run setup-brief --file path/to/cv.txt       # plain text
-//   cat cv.txt | pnpm run setup-brief                # stdin
+//   pnpm run setup-brief --file path/to/cv.pdf           # parse PDF (via pdfjs-dist)
+//   pnpm run setup-brief --file path/to/cv.docx          # parse DOCX (via mammoth)
+//   pnpm run setup-brief --file path/to/cv.md            # plain markdown
+//   pnpm run setup-brief --file path/to/cv.txt           # plain text
+//   pnpm run setup-brief --linkedin path/to/profile.pdf  # LinkedIn "Save to PDF" export
+//   cat cv.txt | pnpm run setup-brief                    # stdin
+//
+// `--linkedin` is the same pipeline as `--file` but tells the LLM the input is
+// a LinkedIn profile export so it ignores LinkedIn's boilerplate. A `--file`
+// whose name contains "linkedin" is auto-treated as a LinkedIn source.
 //
 // Provider: auto-detects claude / codex / gemini / opencode on PATH (in that
 // order). Override with PUPILA_LLM=<provider>.
 
 import { existsSync } from 'node:fs';
 import { copyFile } from 'node:fs/promises';
+import { basename } from 'node:path';
+import { type BriefSource, buildBriefPrompt } from './lib/brief-prompt.js';
 import { writeBriefBody } from './lib/brief-template.js';
 import { detectFormat, parseCvFile } from './lib/cv-parser.js';
 import { detectLlmCli, runLlm } from './lib/llm.js';
@@ -27,11 +34,15 @@ const CV_DEST_BASENAME = 'config/cv';
 
 interface CliArgs {
   file: string | null;
+  source: BriefSource;
   help: boolean;
 }
 
 function parseArgs(argv: string[]): CliArgs {
   let file: string | null = null;
+  // Only flips to 'linkedin' when --linkedin is passed, or when a --file path
+  // looks like a LinkedIn export (see inference below).
+  let source: BriefSource = 'cv';
   let help = false;
   for (let i = 0; i < argv.length; i++) {
     const arg = argv[i];
@@ -44,11 +55,29 @@ function parseArgs(argv: string[]): CliArgs {
         file = next;
         i++;
       }
+    } else if (arg.startsWith('--linkedin=')) {
+      file = arg.slice(11);
+      source = 'linkedin';
+    } else if (arg === '--linkedin' && i + 1 < argv.length) {
+      const next = argv[i + 1];
+      if (next) {
+        file = next;
+        source = 'linkedin';
+        i++;
+      }
     } else if (arg === '--help' || arg === '-h') {
       help = true;
     }
   }
-  return { file, help };
+  // Auto-detect a LinkedIn export passed via --file by its filename, so
+  // `--file ~/Downloads/LinkedIn_Profile.pdf` still gets the tuned prompt.
+  // Match only the basename — a directory like ~/linkedin-stuff/ shouldn't
+  // silently flip a regular CV to the LinkedIn prompt. The "Reading … (…,
+  // LinkedIn export)" log in main() surfaces the decision either way.
+  if (source === 'cv' && file && /linkedin/i.test(basename(file))) {
+    source = 'linkedin';
+  }
+  return { file, source, help };
 }
 
 async function readStdin(): Promise<string> {
@@ -59,21 +88,6 @@ async function readStdin(): Promise<string> {
   return Buffer.concat(chunks).toString('utf-8');
 }
 
-function buildPrompt(cvText: string): string {
-  return `You are summarizing the following CV into a short candidate brief that will be sent to an LLM each time the candidate's job-matching tool evaluates a posting. The brief decides whether the LLM agrees with the rule-based fit score.
-
-Output ONLY three short paragraphs as plain markdown text. No preamble, no markdown fences, no headings, no commentary.
-
-PARAGRAPH 1 — Who they are: role, years of experience, primary location, primary stack/skills. Be concrete (frameworks, languages, tools they ship with regularly).
-PARAGRAPH 2 — What they're looking for: target seniority (senior / lead / staff / principal IC), domains/sectors of interest (web3, AI, fintech, etc.), location preference (remote-worldwide / remote-EMEA / hybrid in <city> / open to relocation).
-PARAGRAPH 3 — What to avoid: roles that look like a fit on paper but aren't. Examples: wrong specialty (backend if frontend, etc.), wrong level (junior, intern, exec), on-site only, US-only positions, support/solutions/devrel/GTM titles.
-
-Aim for 6-10 lines total. Drop anything that doesn't help a job-matching tool decide. Don't editorialize.
-
-CV:
-${cvText.slice(0, MAX_CV_CHARS)}`;
-}
-
 function stripMarkdownFences(text: string): string {
   let cleaned = text.trim();
   if (cleaned.startsWith('```')) {
@@ -89,6 +103,7 @@ async function main(): Promise<void> {
     console.log('  pnpm run setup-brief --file path/to/cv.pdf');
     console.log('  pnpm run setup-brief --file path/to/cv.docx');
     console.log('  pnpm run setup-brief --file path/to/cv.md');
+    console.log('  pnpm run setup-brief --linkedin path/to/profile.pdf   # LinkedIn "Save to PDF"');
     console.log('  cat cv.txt | pnpm run setup-brief');
     console.log('');
     console.log('Provider: auto-detects claude/codex/gemini/opencode on PATH.');
@@ -103,7 +118,8 @@ async function main(): Promise<void> {
       process.exit(1);
     }
     const format = detectFormat(args.file);
-    console.log(`Reading ${args.file} (${format})...`);
+    const sourceLabel = args.source === 'linkedin' ? 'LinkedIn export' : 'CV';
+    console.log(`Reading ${args.file} (${format}, ${sourceLabel})...`);
     try {
       cvText = await parseCvFile(args.file);
     } catch (err) {
@@ -149,7 +165,7 @@ async function main(): Promise<void> {
     `Parsed ${cvText.length} chars. Running ${invocation.provider} (${invocation.cmd})...`,
   );
 
-  const prompt = buildPrompt(cvText);
+  const prompt = buildBriefPrompt(cvText, args.source, MAX_CV_CHARS);
   let raw: string;
   try {
     raw = await runLlm(prompt);

tests/brief-prompt.test.ts

@@ -0,0 +1,44 @@
+import { describe, expect, it } from 'vitest';
+import { buildBriefPrompt } from '../src/lib/brief-prompt.js';
+
+const SAMPLE = 'Jane Doe — Senior Frontend Engineer. React, TypeScript, 8 years.';
+
+describe('buildBriefPrompt', () => {
+  it('keeps the three-paragraph output contract for both sources', () => {
+    for (const source of ['cv', 'linkedin'] as const) {
+      const prompt = buildBriefPrompt(SAMPLE, source, 12_000);
+      expect(prompt).toContain('PARAGRAPH 1 — Who they are');
+      expect(prompt).toContain('PARAGRAPH 2 — What they');
+      expect(prompt).toContain('PARAGRAPH 3 — What to avoid');
+      expect(prompt).toContain('No preamble, no markdown fences');
+      // The raw document text is always appended.
+      expect(prompt).toContain(SAMPLE);
+    }
+  });
+
+  it('labels the input as CV for the cv source', () => {
+    const prompt = buildBriefPrompt(SAMPLE, 'cv', 12_000);
+    expect(prompt).toContain('summarizing the following CV');
+    expect(prompt).toContain('\nCV:\n');
+    // No LinkedIn-specific framing leaks into the CV prompt — check the exact
+    // LinkedIn copy, not just the word, so a partial bleed is caught.
+    expect(prompt).not.toContain('LINKEDIN PROFILE');
+    expect(prompt).not.toContain('Save to PDF');
+    expect(prompt).not.toContain('Ignore LinkedIn boilerplate');
+  });
+
+  it('adds LinkedIn-tuned framing for the linkedin source', () => {
+    const prompt = buildBriefPrompt(SAMPLE, 'linkedin', 12_000);
+    expect(prompt).toContain("summarizing the candidate's LinkedIn profile");
+    expect(prompt).toContain('Save to PDF');
+    expect(prompt).toContain('Ignore LinkedIn boilerplate');
+    expect(prompt).toContain('\nLINKEDIN PROFILE:\n');
+  });
+
+  it('truncates the document text to maxChars', () => {
+    const long = 'x'.repeat(50);
+    const prompt = buildBriefPrompt(long, 'cv', 10);
+    expect(prompt).toContain('xxxxxxxxxx'); // exactly 10
+    expect(prompt).not.toContain('x'.repeat(11));
+  });
+});

ui/plugins/_shared.ts

@@ -20,6 +20,9 @@ export const CV_MAX_CHARS = Number(process.env.PUPILA_CV_MAX_CHARS ?? '12000');
 // `src/types.ts` so the MCP server and the UI agree on the same values.
 export const VALID_STATUSES: ReadonlySet<string> = new Set<string>(APPLICATION_STATUSES);
 export const VALID_CV_FORMATS = new Set<CvFormat>(['pdf', 'docx', 'md', 'txt']);
+// Brief input source — 'cv' (a résumé/CV) or 'linkedin' (a LinkedIn
+// "Save to PDF" export). Only changes the LLM prompt framing.
+export const VALID_CV_SOURCES = new Set<string>(['cv', 'linkedin']);
 export const VALID_PROVIDER_OR_AUTO = new Set<string>([...SUPPORTED_PROVIDERS, 'auto']);
 
 const CV_EXTENSIONS: readonly CvFormat[] = ['pdf', 'docx', 'md', 'txt'];