take2

.claude/dude

@@ -0,0 +1 @@
+/Users/mike/.claude

.claude/skills/cupid0/SKILL.md

@@ -0,0 +1,44 @@
+---
+name: cupid0
+description: Use when searching, filtering, and evaluating job listings for Mike
+---
+
+# cupid0
+
+Job seeker agent. Finds jobs on Indeed, filters rejects, evaluates matches, tailors resume, applies.
+
+## Process
+
+1. **search** — scrape one Indeed page using chrome-devtools MCP (see indeed.md)
+2. **filter** — swipe left based on swipe_left.md, run `node filter.js`
+3. **report** — show keepers as `company | title | salary`, ask Mike before continuing
+4. **next page** — only after Mike gives the go-ahead, repeat steps 1-3
+5. **evaluate** — once all pages done, screen each keeper ONE AT A TIME: fetch full description → save to /tmp/descriptions/<company>.md → score 1-100 vs resume → report → wait → next
+6. **tailor** — generate tailored resume per keeper using ai_golden.md + og_golden.md
+7. **apply** — fill and submit application
+
+## Pace Rules — READ THIS
+
+- **Never parallelize.** Not search, not fetching, not scoring. One thing at a time.
+- After each page: report keepers, stop, wait for Mike.
+- During evaluate: fetch one description, score it, report it, stop, wait for Mike.
+- Natural pauses = natural rate limiting. We have time. Don't rush.
+- Going serial also avoids Indeed detection/rate limits.
+
+## Files
+
+- `swipe_left.md` — auto-reject criteria
+- `swipe_right.md` — green flags that boost score
+- `indeed.md` — Indeed-specific scraping instructions
+
+## Resumes
+
+- `~/dev/self/portfolio/resume/ai_golden.md` — AI/agentic focus
+- `~/dev/self/portfolio/resume/og_golden.md` — full classic experience
+
+## State
+
+- `/tmp/suitors.json` — current candidate list
+- `/tmp/keepers.json` — keepers across pages
+- `/tmp/evals.json` — scores and notes
+- `/tmp/descriptions/` — full job descriptions

.claude/skills/cupid0/indeed.md

@@ -0,0 +1,33 @@
+---
+name: indeed
+description: Use when searching or scraping Indeed job listings for cupid0
+---
+
+# Indeed
+
+## What
+Scrape Indeed job listings using the chrome-devtools MCP against the user's real logged-in Chrome session.
+
+## How
+
+**Search URL:**
+```
+https://www.indeed.com/jobs?q=software+engineer&l=Remote&fromage=7&salaryType=%24150%2C000&sc=0kf%3Aattr%28DSQF7%29%3B
+```
+
+**Pagination:** append `&start=20` for page 2, `&start=40` for page 3, etc.
+
+**Scrape jobs with evaluate_script:**
+```javascript
+() => [...document.querySelectorAll('.job_seen_beacon')].map(el => ({
+  title: el.querySelector('.jobTitle a')?.innerText?.trim(),
+  company: el.querySelector('[data-testid="company-name"]')?.innerText?.trim(),
+  url: el.querySelector('.jobTitle a')?.href
+})).filter(j => j.title)
+```
+
+**Rules:**
+- Use chrome-devtools MCP only — Playwright gets detected on page 2+
+- Do NOT navigate if already on the right page — read first
+- After getting jobs from evaluate_script, use Write tool to save to /tmp/cupid0_jobs.json as {"count": N, "jobs": [...]}
+- Run `node filter.js` after each page

.claude/skills/cupid0/swipe_left.md

@@ -0,0 +1,25 @@
+# swipe left 🙅
+
+auto-reject any job that matches these. no need to fetch the description.
+
+- government, federal, or public sector
+- military, defense, army, dod, or requires security clearance
+- religious organizations
+- crypto, blockchain, web3, nft, solana
+- staffing agencies / body shops (Intone, Wipro, Infosys, TCS, etc.)
+- consulting farms that just place you at clients (Accenture, Deloitte, etc.)
+
+## tech i don't do
+
+- crypto / blockchain / web3
+- android / mobile
+- java
+- golang
+- devops / sre / platform engineering
+- cybersecurity / infosec
+- php
+
+## company stage
+
+- pre-revenue startups (no salary, equity-only)
+- solo engineer / first engineer at non-tech companies (CPA firms, law firms, etc.)

.claude/skills/cupid0/swipe_right.md

@@ -0,0 +1,12 @@
+# swipe right 🙌
+
+auto-bump score for any job mentioning these. rare green flags.
+
+## culture
+
+- XP (Extreme Programming)
+- pair programming
+- mob programming
+- TDD / BDD / ATDD
+- Kanban
+- DDD (Domain-Driven Design)

.gitignore

@@ -21,3 +21,5 @@ erl_crash.dump
 
 # Temporary files, for example, from tests.
 /tmp/
+/.claude/dude/
+/.idea/

.mcp.json

@@ -0,0 +1,8 @@
+{
+  "mcpServers": {
+    "cupid0": {
+      "command": "npx",
+      "args": ["chrome-devtools-mcp@latest", "--autoConnect"]
+    }
+  }
+}

CLAUDE.md

@@ -0,0 +1,21 @@
+# cupid0
+
+## FULL DESCRIPTIONS ONLY. PERIOD.
+
+When showing a job description, ALWAYS paste the FULL raw text — every word, no summarizing, no bullets, no "here's what it says".
+If it's saved in /tmp/descriptions/, cat the whole file.
+If it's not saved, fetch and paste everything.
+NO EXCEPTIONS. NO SUMMARIES. NO REFORMATTING.
+
+## SHORTCUTS
+
+- desc: paste the entire file content, every word, no summarizing, no reformatting, no bullets
+
+## TAILORING RULES
+
+- Source: og_golden.md for style and impact bullets, ai_golden.md for AI/agentic work
+- Keep og's precise impact format — no vibe fluff
+- Where og doesn't cover what the job asks, pull from ai_golden but rewrite in og's style
+- Only go back far enough in history to fill the role — no need to list everything
+- Select and remix the best matching bullets, don't copy-paste everything
+- Output is one tight resume that reads like og but speaks the job's language