daily-job-scan

name	daily-job-scan
description	Daily job scan — find new postings, score them, build custom resumes for roles scoring 6+

You are running an automated daily job scan and resume generation task for [YOUR FULL NAME] ([YOUR EMAIL]), a [SENIORITY LEVEL] [YOUR FIELD] professional actively searching for a new position.

Each run: find new job postings in your target areas, score them against fit criteria, generate a custom resume for every role scoring 6+, and present a full report including both passing and failing roles.

---

STEP 1: LOAD CONTEXT

Read these files before doing anything else:

• ~/Claude-Cowork-Workspace/context/about-me.md — full career history, certifications, skills
• ~/Claude-Cowork-Workspace/CLAUDE.md — resume building rules (mandatory)
• ~/Claude-Cowork-Workspace/memory/context/job-pipeline.md — current applications (for de-duplication)
• ~/Claude-Cowork-Workspace/memory/working-memory.md — compensation context, geography rules

---

STEP 2: SEARCH FOR NEW ROLES

Use Claude in Chrome (mcp__Claude_in_Chrome__tabs_context_mcp to get or create a tab) for LinkedIn searches. Run all searches below and collect every unique result.

LOCAL SEARCHES — Within [X] miles of [YOUR CITY], all work types, posted last 24 hours:

Search A — [PRIMARY ROLE TYPE]: https://www.linkedin.com/jobs/search/?keywords=[KEYWORDS_A]&location=[YOUR_CITY_ENCODED]&distance=[MILES]&f_TPR=r86400&sortBy=R

Search B — [SECONDARY ROLE TYPE]: https://www.linkedin.com/jobs/search/?keywords=[KEYWORDS_B]&location=[YOUR_CITY_ENCODED]&distance=[MILES]&f_TPR=r86400&sortBy=R

Search C — [TERTIARY ROLE TYPE]: https://www.linkedin.com/jobs/search/?keywords=[KEYWORDS_C]&location=[YOUR_CITY_ENCODED]&distance=[MILES]&f_TPR=r86400&sortBy=R

NATIONAL REMOTE SEARCHES — Remote only (f_WT=2), posted last 24 hours:

Search D — [REMOTE ROLE TYPE 1]: https://www.linkedin.com/jobs/search/?keywords=[KEYWORDS_D]&f_WT=2&f_TPR=r86400&sortBy=R

Search E — [REMOTE ROLE TYPE 2]: https://www.linkedin.com/jobs/search/?keywords=[KEYWORDS_E]&f_WT=2&f_TPR=r86400&sortBy=R

Search F — [REMOTE ROLE TYPE 3]: https://www.linkedin.com/jobs/search/?keywords=[KEYWORDS_F]&f_WT=2&f_TPR=r86400&sortBy=R

Also run a WebSearch: "[YOUR KEY TITLE KEYWORDS]" site:boards.greenhouse.io OR site:jobs.lever.co OR site:myworkdayjobs.com remote 2026

CUSTOMIZATION NOTE: Replace [KEYWORDS_X] with URL-encoded search terms. Spaces become %20, quotes become %22. Examples:

• "security architect" → security%20architect
• "head of engineering" → %22head%20of%20engineering%22

De-duplicate all results. Remove clearly off-lane titles (define your own exclusions below):

• [EXCLUSION 1 — e.g., "Analyst (non-senior, non-architect)"]
• [EXCLUSION 2 — e.g., "Recruiter / HR roles"]
• [EXCLUSION 3 — e.g., "Sales / pre-sales / solutions engineer"]

Check ~/Claude-Cowork-Workspace/Resume/Hunt/Applied Companies/ — if a company subfolder already exists, note "ALREADY APPLIED" but still include the role in scoring.

---

STEP 3: SCORE EACH ROLE (1–10 scale)

Start at 5. Adjust up or down:

Role type: +3 = [YOUR TOP TARGET ROLE — e.g., VP Engineering, Head of Product, Principal Architect] +3 = [SECOND TOP TARGET — e.g., Director at growth-stage company] +2 = [GOOD FIT ROLE — e.g., Staff-level IC, Senior Director at enterprise] +1 = [DECENT FIT — e.g., Senior Manager with path to Director] -1 = [SLIGHT MISMATCH — e.g., Consulting/advisory role] -2 = [SIGNIFICANT MISMATCH — e.g., Manager without Director path] -3 = [WRONG LEVEL — e.g., IC contributor, no leadership scope]

Fit signals: +1 = [PREFERRED INDUSTRY 1 — e.g., FinTech, AI/ML, SaaS] +1 = [PREFERRED INDUSTRY 2] +1 = [LOCAL BONUS — within [Y] miles of home] +1 = [YOUR KEY CREDENTIAL explicitly listed as preferred]

Red flags (reduce score): -2 = [HARD SKILL GAP 1 — e.g., requires specific tool you don't have] -2 = [INDUSTRY MISMATCH — e.g., automotive, hardware] -2 = Posted comp range tops out below $[YOUR_COMP_FLOOR] base (flag as COMP MISS)

Cap at 10, floor at 1.

Record score AND primary reason for score (one phrase) for every role, regardless of score.

---

STEP 4: FOR EACH ROLE SCORING 6+, GENERATE A CUSTOM RESUME

Do this for every 6+ role, in sequence.

4a. Get the full job description, verify the posting is active, and capture the application URL

Navigate to the job posting URL in Chrome and use get_page_text to extract the full JD.

Verify the posting is still active before building a resume. A posting is NOT active if:

• The page returns a 404 or "job no longer exists" message
• The page says "No longer accepting applications"
• The ATS page is empty or redirects to a generic jobs page

If the posting is inactive, skip resume generation and note "POSTING EXPIRED" in the report.

Capture the direct application URL — required in the final report. Priority:

1. The "Apply" button destination (Greenhouse, Lever, Workday, company ATS)
2. The job posting page URL itself
3. LinkedIn job view URL if company is confidential

4b. Read base resumes

# Use pandoc to extract text from each base resume
pandoc ~/Claude-Cowork-Workspace/Resume/[YourName]_Resume_[Variant1].docx -t plain
pandoc ~/Claude-Cowork-Workspace/Resume/[YourName]_Resume_[Variant2].docx -t plain

4c. Synthesize and build resume

Select the most appropriate base resume for the role type. Tailor it to the JD by:

• Adjusting the summary to match the role's language and focus
• Reordering bullet points to lead with most relevant accomplishments
• Mirroring key terms from the JD (for ATS matching)
• Adjusting the title line to match the role's language

Build the resume as a Node.js docx-js script:

• NODE_PATH=/usr/local/lib/node_modules_global/lib/node_modules
• US Letter: width 12240, height 15840 DXA; margins 1080 DXA all sides
• Font: Arial, 20pt body (= 10pt — docx size is in half-points)
• NEVER use Unicode bullets — use LevelFormat.BULLET with Word numbering config, text: "-"
• No tables, no headers, no footers
• Tab stops for right-aligned dates: position 10080
• Section headers: bold 22pt (= 11pt) navy (#1F3864) with bottom border
• Title line matches JD role title language
• Contact line: [YOUR_PHONE] | [YOUR_EMAIL] | linkedin.com/in/[YOUR_LINKEDIN]
• Use "--" instead of em dashes
• Output: ~/Claude-Cowork-Workspace/Resume/[YourName]Resume{CompanyName}_{RoleAbbrev}.docx

Run and validate:

cd ~/Claude-Cowork-Workspace/outputs && NODE_PATH=/usr/local/lib/node_modules_global/lib/node_modules node build_resume_{CompanyName}.js
VALIDATE_SCRIPT=$(find /sessions/*/mnt/.claude/skills/docx/scripts/office/validate.py 2>/dev/null | head -1)
python3 "$VALIDATE_SCRIPT" ~/Claude-Cowork-Workspace/Resume/[YourName]_Resume_{CompanyName}_{RoleAbbrev}.docx

4d. Cover letter

For roles scoring 8+, also generate a brief cover letter (.docx). Lead with the strongest fit narrative for that role type. Save as: [YourName]CoverLetter{CompanyName}_{RoleAbbrev}.docx

---

STEP 5: PRESENT RESULTS

Use mcp__cowork__present_files to present each generated resume and cover letter.

Then output the full report in this exact format:

---

Daily Job Scan -- {YYYY-MM-DD}

Scanned: {N} total roles | Scoring 6+: {N} | Below threshold: {N} | Resumes generated: {N}

---

ROLES SCORING 6+ (sorted highest to lowest)

[Score]/10 -- [Role Title] | [Company] | [City or Remote] Comp: {posted range or "not listed"} | {On-site/Hybrid/Remote} Apply: {direct URL} Resume: {file path} {Cover letter path if generated} Fit note: {2-3 sentences on fit, any flags} {ALREADY APPLIED if applicable}

---

ROLES BELOW THRESHOLD (sorted by score, highest to lowest)

Score	Role	Company	Location	Why it didn't make it

(Include ALL roles scanned that scored 1-5, with one-liner reason for each)

---

If no roles score 6+:

No new 6+ roles found today. See below-threshold list for all {N} roles scanned.

---

IMPORTANT CONSTRAINTS

• Do NOT apply to any roles — scan, score, and generate only
• Do NOT modify job-pipeline.md during this task — update the pipeline manually
• If Chrome is unavailable, fall back to WebSearch for all searches and note the limitation
• If docx generation fails for a role, note the error and continue with remaining roles
• Each run is independent — do not reference prior scan results
• f_TPR=r86400 (last 24 hours) is intentional — this is a daily scan, not a backfill