القائمة
Two modes: (1) full job scan with deep warm-path detection, and (2) warm-path enrichment on existing scan results. Reads targeting from config, scans LinkedIn + ATS, fetches JDs, detects 1st/2nd-degree warm paths, and writes scan reports + pipeline updates.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use this skill whenever the user asks to implement, build, fix, ship, add, refactor, migrate, or change ANY code — features, bugs, plugins, skills, migrations, hooks, rules, configs, in any repo. Triggers on 'build X', 'fix Y', 'implement Z', 'ship this to <plugin>', a ticket id with code work. This is the Agentic SDLC pipeline; do NOT hand-roll `codex exec` or ad-hoc git/PR shipping — route through Stage 0 (claim), the 9 core stages, the Gate-A.5 change-manifest gate, Gate-A.7 design-review gate, Stage 5.5/5.6/5.7/5.8 quality gates, and Stage 10 (completion).
Activate real-time prompt sharpening and persona detection. Use when the user says "co-dialectic", "cod", "codi on", "improve my prompts", "prompt sharpening", "teach me to prompt", or wants to improve their AI communication skills. Provides status line, persona system, caliber enforcement, prompt improvement, context management, auto-codification protocols, and auto-verification by stakes (Protocol 8 — fires hallucination-detector / judge-panel / unknown-unknown / canonical-claim verifier automatically before every substantive output, scaled to the stakes of the artifact).
Surfaces one-week and two-week follow-up nudges for applications that are applied and awaiting response, with a plain-text draft ready for the human to approve. Draft-only: never sends. Use when the user asks what to follow up on, stale applications, application nudges, or job-search follow-ups.
Detects explicitly shareable career milestones in the job pipeline and surfaces draft-only brand-post prompts. Fail-closed by firewall design: no shareable flag means no suggestion. Use when the user asks for milestone brand moments, career-to-brand prompts, or shareable pipeline milestones.
Career OS home screen and central router. Two jobs: show the dashboard and route every action to the correct skill. Mission Control reads and presents but never directly edits persistence files.
Maps your network to target companies. Discovers warm intro paths, scores connection strength (5-level scale), ingests new contacts from conversation, and recommends who to reach out to and how. Also detects relationship origin (where/when you met) and cohort clusters (people from the same company/era). Records inbound recruiter DMs that came from the user's public posts as brand-attributed pipeline opportunities. Say "who do I know at [Company]", "how do I know [Name]", "tell me about [Contact]", or "a recruiter DM'd me after seeing my post".
| name | job-search-scheduler |
| description | Two modes: (1) full job scan with deep warm-path detection, and (2) warm-path enrichment on existing scan results. Reads targeting from config, scans LinkedIn + ATS, fetches JDs, detects 1st/2nd-degree warm paths, and writes scan reports + pipeline updates. |
| triggers | ["scan for jobs","js","job search","find matching roles","what jobs match me","scan for jobs today","daily job scan","any new roles","find new roles","run job scan","what's new on the market","enrich warm paths","warm paths","find connections","who knows people at these companies","check my network","find referrals","warm intros for scan","network scan","enrich contacts","find warm leads","run warm path detection","check connections","linkedin people check"] |
$CAREER_GITHUB_REPOis derived from:git -C $CAREER_HOME remote get-url origin | sed 's/.*github.com[:/]//;s/.git$//'
Tasks live in $CAREER_GITHUB_REPO GitHub Issues (canonical source of truth — single inbox for all Cyborg work). Repo of work indicated by repo:* label, NOT by issue location. Cadence indicated by cadence:* label (operational for high-frequency churn; strategic for sprint-scale; meta for trackers). Tier indicated by tier:* label (p1/p2/p3/backlog).
Tasks.md is DEPRECATED as of v0.25.0. See $CAREER_HOME/workspace.manifest.yaml task_routing: section for the full architecture.
This skill reads/writes via:
gh CLI (universal, all agents)github-mcp MCP server (post-restart, when MCP boots — at npx @modelcontextprotocol/server-github)New role discoveries open as kind:scan-result issues (one per high-priority finding) so cruise-control / mission-control can pick them up:
gh issue create --repo $CAREER_GITHUB_REPO \
--title "Score+apply: {Company} {Role}" \
--label "tier:p2,cadence:operational,repo:career-os-data,kind:scan-result" \
--body "<scan-source URL + JD path + warm-path tag + suggested resume track>"
Scan reports themselves remain markdown files at career-intelligence/projects/job-search/scans/{date}/scan-{HH}-{MM}.md — those are READ artifacts, not tasks.
Prerequisite: career-intelligence-onboarding must be complete.
Say "onboard me to career intelligence" to run it. It creates brain/identity/experience-history.md and career-intelligence/projects/job-search/job-search-config.md from an 8-question interview. Without these files, the scanner has no targeting config and will produce irrelevant results.
Step 1 — Expand job search config
After onboarding, say: "set up my job search config" or "expand job search targeting". This reads your preferences from job-search-config.md and creates the full operational config at brain/config/job-search.md — adding ATS direct URLs, LinkedIn search keywords, company tiers, and warm-path settings. This is what the scanner reads on every run.
Step 2 — Add your resume tracks
Create a folder Resumes & Cover Letters/ in your workspace. Paste or upload your current resume as one or more files. The file names become your track names (e.g., resume-engineering-leader.md, resume-executive.md). The resume engine auto-detects tracks from filenames. If you have only one resume, name it resume-base.md — the engine will derive tracks from your experience history.
Step 3 — Set up the 6 AM automated scan (optional) Use Claude Code's scheduled task system to run the scanner automatically:
0 6 * * 1-5 (6 AM Monday–Friday)skills/job-search-scheduler/scan-prompt-v11.md (everything below the ## PROMPT BODY line)$CAREER_HOME directory
The automated scan runs in surface-level mode (no deep People tab navigation). You review the output and run "enrich warm paths" manually for high-priority roles.What to customize per your situation:
brain/config/job-search.md → edit Target Roles, Filters, ATS Direct URLs, LinkedIn Search KeywordsTwo jobs:
The scan produces a role list. The enrichment produces actionable referral paths. Both write to the same scan directory. Together they form the highest-ROI step in the job search: finding roles AND finding the people who can get you in.
Always start your response with:
━━━ Career OS: Job Search ━━━
For enrichment mode:
━━━ Career OS: Warm Path Enrichment ━━━
Say: "scan for jobs", "any new roles?", "daily job scan", "find matching roles", "what's new on the market", "run job scan"
Runs the complete pipeline: discover → JD fetch → quick filter → warm-path detection → write scan report → pipeline write-back.
Say: "enrich warm paths", "warm paths", "find connections", "check my network", "find referrals", "warm intros for scan", "network scan", "enrich contacts", "find warm leads", "linkedin people check", "check connections"
Reads the most recent scan report, identifies roles with Cold or surface-level warm tags, and runs deep LinkedIn People tab navigation for 1st-degree, 2nd-degree, and alumni network detection. Writes an enrichment addendum to the same scan date folder.
Runs via daily-linkedin-job-scan scheduled task. Surface-level warm tagging
only (no People tab navigation — too slow for headless runs). Writes scan
report for user to review or enrich manually.
| Artifact | Location | Relationship |
|---|---|---|
| Config (targeting) | brain/config/job-search.md | Source of truth for all targeting parameters |
| Scheduled task prompt | skills/job-search-scheduler/scan-prompt-v11.md | Canonical scanner prompt (v11 — plugin path format). Paste into scheduled-task config. |
| Mission-control routing | skills/mission-control/SKILL.md | Routes scan + enrich triggers to this skill |
| Job-match-scorer skill | skills/job-match-scorer/SKILL.md | Receives scan output for deep scoring |
| Network-intelligence skill | skills/network-intelligence/SKILL.md | Deep contact analysis for individual companies |
If you change targeting → edit config only (both consumers read it). If you change scan logic → update this SKILL.md AND the scheduled task prompt.
| Source | Path | What It Provides |
|---|---|---|
| Config | brain/config/job-search.md | Target roles, company tiers, ATS URLs, LinkedIn keywords, filters, skip rules, warm-path settings |
| JD samples | brain/reference/jd-samples/*.md | Saved JD snapshots for roles already fetched |
| Pipeline | career-intelligence/projects/job-search/job-pipeline.json | Already Applied table (skip dupes) + Warm Intros table (existing contacts) |
| Skills matrix | brain/identity/skills-matrix.md | Technology proficiency for quick-filter gap detection |
| People | network/people/*.md | Contact network for warm-path cross-referencing |
| Identity | brain/identity/identity.md | Values, philosophy for culture fit |
| LinkedIn (via Chrome MCP) | Browser navigation | Job feeds, company pages, People tabs, connection data |
| Previous scan | career-intelligence/projects/job-search/scans/{latest}/*.md | Delta tracking — what changed since last scan |
| Output | Path | What It Contains |
|---|---|---|
| Scan report | career-intelligence/projects/job-search/scans/{YYYY-MM-DD}/scan-{HH}-{MM}.md | Roles found, warm tags, JD quality, verify queue |
| Enrichment report | career-intelligence/projects/job-search/scans/{YYYY-MM-DD}/enrich-{HH}-{MM}.md | Deep warm-path results for roles from latest scan |
| JD snapshots | brain/reference/jd-samples/{company}-{role-slug}.md | Fetched JD text (permanent — URLs die, snapshots don't) |
| Task updates | GitHub Issues $CAREER_GITHUB_REPO | New kind:scan-result issues opened per high-priority role |
| Pipeline updates | career-intelligence/projects/job-search/job-pipeline.json | New "Ready to Apply" entries |
When the skill is invoked, determine mode from the trigger:
| Trigger matches | Mode | Action |
|---|---|---|
| Scan triggers ("scan for jobs", "any new roles", etc.) | Full Scan | Run complete pipeline (Steps 1-5) |
| Enrich triggers ("enrich warm paths", "warm paths", etc.) | Enrichment Only | Skip discovery, run warm-path enrichment on latest scan (Step 3 only) |
| "scan and enrich" or "full scan with warm paths" | Full Scan + Enrichment | Run complete pipeline including deep warm paths |
Read from config file (brain/config/job-search.md):
Read pipeline + people files:
Read most recent scan file → for delta tracking.
If Chrome MCP unavailable: skip Source A entirely. Log it. Run Source B only.
A1. Recommended feed: Navigate linkedin.com/jobs → capture first 20 matching roles from "Recommended for you."
A2. Keyword searches: Run each keyword from config's LinkedIn Search Keywords
section at linkedin.com/jobs/search/ with filters from config's
LinkedIn Search Filters section. First 15 results per search. Dedup.
A3. Connection noting: While on each LinkedIn job page, if LinkedIn shows "X connections at this company" or names 1st-degree contacts, note them. Do NOT navigate to People tabs in this step (that's Step 3).
Fallback: If Chrome MCP unavailable or LinkedIn logged out, skip and log.
WebFetch each URL from config's ATS Direct URLs table. Look for role titles
matching target roles from config. WebFetch is faster and cheaper than Chrome
for server-rendered ATS pages.
JS-rendered sites from config's JS-Rendered Sites list are covered by
LinkedIn in Source A only.
For every NEW role found:
brain/reference/jd-samples/, tag ✅ JD🔄 partial⚠️ title-onlyJD URL validation rules:
| URL Pattern | Valid? | Action |
|---|---|---|
https://jobs.ashbyhq.com/... | Yes | Fetch JD |
https://job-boards.greenhouse.io/... | Yes | Fetch JD |
https://jobs.lever.co/... | Yes | Fetch JD |
https://careers.{company}.com/... | Yes | Fetch JD |
https://linkedin.com/company/... | No | Company page — click through |
https://linkedin.com/jobs/search/... | No | Search results — not a JD URL |
Rescore notification (WO-035): After fetching JDs, check if any newly-fetched JD
matches a role already in the match tracker with ⚠️ title-only Quality. If so,
append a note to the scan report:
⚠️ {N} roles now have JDs that were previously scored title-only — run "rescore queue" to update scores
This is a notification only — do not auto-trigger rescoring (token cost is Opus-level, user decides when to spend it).
For every role found, check company name against:
Tag with strongest match (priority: 🔥 ACTIVE > ✅ REFERRAL > ✅ Resume sent > other). Note LinkedIn connection counts from A3.
This is the same level the automated 6 AM scan does. Deep detection is Step 3.
This step only runs when the user invokes the skill manually (not the scheduled task).
For every NEW role found with ≥80% quick filter score, extract connection data from the company's LinkedIn People tab.
Primary method: DOM/JS extraction (10-20x cheaper than screenshots)
Navigate directly to the People tab URL:
https://www.linkedin.com/company/{slug}/people/
This skips the "navigate to company page → click People tab" round-trip.
Extract via JavaScript (javascript_tool or equivalent JS execution):
Query DOM for people cards — name, title, connection degree, mutual connections.
Do NOT take screenshots to visually parse the page.
Filter via URL parameters instead of click-filter-screenshot:
/company/{slug}/people/?facetNetwork=%5B"F"%5D/company/{slug}/people/?facetNetwork=%5B"S"%5D/company/{slug}/people/?keywords=BerkeleyRate limiting: Wait 2-3 seconds between companies to avoid LinkedIn throttling.
Fallback: If JS extraction fails (empty results, CAPTCHA, DOM change):
network/people/*.mdnetwork/people/*.md → actionable intro path2nd-degree scoring:
| Mutual Connection Type | Strength | Action |
|---|---|---|
| In people/ with 🔥 ACTIVE status | Very strong | Ask mutual for direct intro |
| In people/ with any status | Strong | Reach out to mutual, mention shared connection |
| Berkeley Haas EMBA cohort | Strong | Alumni warm outreach |
| LinkedIn 1st but not in people/ | Medium | Connect first, then ask |
| Poker gang / social circle | Very strong | Text/WhatsApp ask |
?keywords=Berkeley (or "Haas")For roles scoring ≥85% with warm paths found, suggest:
"who do I know at [Company]" → routes to network-intelligence for deep
contact analysis and outreach draft.
Apply quick filter from config to disqualify obvious mismatches. Deep 6-category scoring is handled by the job-match-scorer skill.
| Factor | Weight |
|---|---|
| Role level fit | 20% |
| Company tier | 10% |
| Recency | 5% |
| Salary fit | 5% |
If quick filter < 50% → SKIP.
After scan completes, suggest: "score the latest scan" to invoke job-match-scorer.
Scan report (career-intelligence/projects/job-search/scans/{YYYY-MM-DD}/scan-{HH}-{MM}.md):
# Job Scan — {date} {time}
## Search Mode: {from config}
**Candidate:** {read from brain/identity/experience-history.md → `who:` frontmatter field}
**Scan type:** Manual scan with deep warm-path detection
**Reference:** Previous scan {date} {time}
---
## Pre-Scan Check
Already Applied: {N} rows counted (header said {M})
Last scan: {date} {time} — {path}
People files: {N} contacts loaded
---
## Summary
| Source | Scanned | New | Skipped | Dead |
|--------|---------|-----|---------|------|
## New Roles
| # | Company | Title | Location | Track | Warm Path | JD Quality | Source | URL |
## Warm Paths Found
| Company | Role | Contact | Type | Strength | Action |
|---------|------|---------|------|----------|--------|
## 2nd-Degree Intro Paths (HIGH VALUE)
| Company | Role | 2nd-Degree Contact | Mutual Connection | Strength | Action |
|---------|------|--------------------|-------------------|----------|--------|
## Alumni Networks
| Company | Alumni Count | Notable Names | Action |
## New Companies (not on ATS list)
| Company | Roles | Stage | Add to list? |
## Status Changes (vs last scan)
| Company | Title | Change |
## Verify Queue
| Company | Role | Source URL | JD Quality | Failure Reason | Action |
## No Matching Roles
| Company | Notes |
## Skipped (with reason)
| Company | Role | Reason |
## Next Steps
→ "score the latest scan" — run Opus scoring on new roles
→ "enrich warm paths" — run deep warm-path detection (if not already done)
→ "who do I know at [Company]" — deep network analysis for a specific company
→ "dashboard" — see apply-ready queue
Pipeline write-back:
Task updates:
kind:scan-result issues on $CAREER_GITHUB_REPO for high-priority new roles (one issue per role; see Task Substrate section above for command)When user says "enrich warm paths", "find connections", "check my network", etc.:
career-intelligence/projects/job-search/scans/ (any date directory, pattern scan-*.md)enrich-*.md in the same date directoryScan enriched: field references itIf no scans exist: "No scan found. Say 'scan for jobs' to run one first." If all scans already enriched: "All scans have been enriched. Run a new scan first."
For each role from the unenriched scans that needs enrichment (Cold or surface-level only):
If Chrome MCP unavailable: "Chrome MCP needed for LinkedIn People tab navigation. Run this in a session with Chrome access."
Save to career-intelligence/projects/job-search/scans/{YYYY-MM-DD}/enrich-{HH}-{MM}.md:
# Warm Path Enrichment — {date} {time}
**Scans enriched:** {scan-09-00.md, scan-10-30.md, ...}
**Roles analyzed:** {N total, M after deduplication}
**New warm paths found:** {N}
**2nd-degree intros found:** {N}
## Warm Path Results
### {Company} — {Role}
| Path | Type | Strength | Action |
|------|------|----------|--------|
| {Name} (1st, {Title}) | Direct referral | Very strong | Ask to refer |
| {Name} (2nd via {Mutual}) | 2nd-degree intro | Strong | Ask {Mutual} for intro |
| {N} Haas alumni | Alumni network | Medium | Alumni outreach |
## Priority Actions (sorted by strength)
| # | Company | Contact | Action | Strength |
|---|---------|---------|--------|----------|
## Still Cold (no paths found)
| Company | Role | LinkedIn Connections | Suggestion |
|---------|------|---------------------|------------|
→ "write outreach for [Name]" — draft intro request
→ "who do I know at [Company]" — deeper analysis on specific company
If the original scan report has warm-path tags, update them with enriched results. Use surgical edits — do not rewrite the entire scan file.
When brain/config/job-search.md does NOT exist:
Walk through setup questions (one at a time, P11):
Create config file and register scheduled task via schedule skill.
User can request targeted scans:
On weekly cadence or when user says "find new companies":
When dashboard loads, check for today's scan: