Career-Ops Skills: Complete Comparison of All 12 Modes (2026)

Career-Ops is one of the most elaborate Claude Code skill suites built to date. It ships 12 distinct modes that handle everything from job board scraping to ATS-optimized CV generation to interview prep. But the 12-mode design is dense — figuring out which mode to use for which task is non-obvious from the README alone.

This reference compares all 12 Career-Ops skill modes side-by-side: what each does, when to use it, output format, and how they chain together in typical workflows. Based on the official santifer/career-ops repository as of May 2026.

The 12 Career-Ops Modes at a Glance

Mode 1: Default — The Entry Point

Trigger: Just run /career-ops with no args, or open the workspace and Claude reads the default mode automatically.

What it does: Lists all 12 modes with a one-line description each, plus quick examples.

When to use: You forgot what modes exist, or onboarding a new user.

This is also the mode Claude reads when starting a session, so the default file effectively serves as the project’s CLAUDE.md-equivalent for the Career-Ops skill suite.

Mode 2: Auto-pipeline — The “Just Make It Happen” Mode

Trigger: Paste a job URL or paste the full job description text.

What it does: Runs the full pipeline:

1. Parses URL or JD text
2. Extracts requirements, salary, location, seniority
3. Runs A-F evaluation across 10 weighted dimensions
4. If score ≥ 4.0/5: generates ATS-optimized CV PDF
5. Adds entry to tracker board with status “evaluated”
6. Generates personalized outreach message (optional)

Output: Evaluation report (markdown) + CV (PDF) + tracker entry.

When to use: You found one specific opportunity and want everything done in one shot. This is the most common workflow.

Time investment: ~3-5 minutes per opportunity.

Mode 3: Scan — Job Board Blitz

Trigger: /career-ops scan with optional filters (location, seniority, remote, etc.).

What it does: Scrapes job listings across 45+ supported companies and job portals:

• Direct careers pages (Stripe, Anthropic, Vercel, Linear, etc.)
• Job aggregators (Y Combinator Jobs, AI Grant, etc.)
• Stage-filtered (Series A/B/C, public companies)

Output: Pre-filtered list of 10-50 opportunities matching your filters, each tagged with company stage, salary range estimate, and a quick fit score.

When to use: Active job search, weekly market scan, or scoping a market segment.

Important: Scan does NOT run full A-F evaluation — that’s reserved for Batch or Auto-pipeline. Scan is the “wide net” before the “deep dive.”

Mode 4: PDF — ATS-Optimized CV Generation

Trigger: /career-ops pdf <job-description>

What it does:

1. Reads your master cv.md file
2. Parses the job description for ATS-relevant keywords
3. Reorders skills, projects, and experiences to maximize keyword match
4. Generates a tailored PDF with proper margins, fonts, and section headers
5. Injects 80% of the JD’s high-frequency keywords into the CV naturally

Output: Single-page or two-page PDF, ATS-parseable, ready to upload.

When to use: After deciding to apply to a specific role.

Key feature: Avoids keyword stuffing. The system uses semantic placement so the CV reads naturally to humans while passing ATS keyword filters.

Mode 5: Batch — Parallel Evaluation

Trigger: /career-ops batch <batch-file.md>

What it does: Spawns multiple claude -p worker processes orchestrated via batch-runner.sh. Each worker:

1. Receives a self-contained prompt from batch/batch-prompt.md
2. Evaluates one opportunity in parallel
3. Returns score + key insights

Output: Scored evaluation table (markdown) ranking all opportunities in the batch file.

When to use: You have 10+ opportunities to evaluate quickly. Batch can score 10 offers in ~5 minutes (vs ~30 minutes serially).

Technical note: This is a textbook example of the parallel agent pattern — a lead agent (your interactive Claude Code session) coordinates with worker agents (the claude -p processes). Each worker has independent context, so cross-contamination is impossible.

Mode 6: Tracker — Status Dashboard

Trigger: /career-ops tracker

What it does: Reads the tracker JSON/markdown and displays a status board:

• Evaluated — opportunities scored but not yet applied to
• Applied — applications submitted
• Screening — recruiter screen scheduled or completed
• Interview — technical interview pipeline
• Offer — offer received
• Rejected / Withdrawn — closed
• Stale — no movement in 14+ days

Output: Kanban-style dashboard with counts per stage and oldest-in-stage warnings.

When to use: Weekly check-in to spot stalled applications and prioritize follow-ups.

Mode 7: Apply — Form-Fill Assistance

Trigger: /career-ops apply <application-form-url>

What it does:

1. Reads the application form structure (questions, character limits)
2. Drafts answers based on your cv.md, the role description, and prior STAR stories accumulated in data/interview-prep/
3. Returns suggested answers for each form field

Output: Form-by-form answer suggestions (NOT auto-submission).

When to use: Long application forms with custom questions (“Why this company?”, “Describe a project you led that failed”).

Important: Career-Ops emphasizes human-in-the-loop. The mode generates suggestions; you submit applications manually. This is deliberate — auto-submission via headless browser is supported by some tools but Career-Ops explicitly avoids it.

Mode 8: Pipeline — Queue Processor

Trigger: /career-ops pipeline

What it does: Reads data/pending-urls.md (a queue of opportunities you’ve copied but not yet evaluated) and runs Auto-pipeline on each one sequentially.

Output: Evaluation reports + CV PDFs + tracker entries for the queue.

When to use: End-of-week batch processing, when you’ve collected 5-10 URLs during the week and want them all evaluated overnight.

Contrast with Batch: Pipeline is serial (one at a time), Batch is parallel (multiple at once). Pipeline is better for ≤10 opportunities; Batch shines at 10+.

Mode 9: Contacto — LinkedIn Outreach

Trigger: /career-ops contacto <linkedin-profile-url>

What it does:

1. Parses the LinkedIn profile (company, role, mutual connections, recent posts)
2. Drafts a personalized outreach message in your preferred tone
3. Includes a specific reference point (their recent post, shared connection, mutual interest)

Output: 1-3 message variants ready to copy-paste.

When to use: Reaching out to hiring managers, internal referrers, or peers at target companies.

Anti-pattern protection: The mode refuses to generate generic templates. If the target profile has insufficient public info, the mode requests more context rather than fabricating personalization.

Mode 10: Deep — Company Research

Trigger: /career-ops deep <company-name>

What it does: Compiles a comprehensive brief:

• Recent funding rounds and stage
• Key product launches in the past 12 months
• Engineering team size and composition (from LinkedIn data)
• Glassdoor sentiment summary
• Recent press coverage
• Tech stack inferences (from job postings, GitHub orgs, public APIs)
• Cultural signals (blog posts, podcasts, public statements)

Output: 1-2 page company brief in markdown.

When to use: Before a technical interview, to walk into the conversation with specific context. Or before a big decision (offer comparison, accepting/declining).

Time investment: ~5-10 minutes for a thorough brief.

Mode 11: Training — Course/Certification Evaluation

Trigger: /career-ops training <course-url-or-name>

What it does: Assesses the ROI of a specific course or certification:

• Cost (time + money)
• Market value (does this keyword appear in target job descriptions?)
• Comparable alternatives (free or cheaper)
• Recency check (is this course based on current tech, or outdated?)
• Conversion fit (does this fit your stated career direction?)

Output: Go/no-go recommendation with reasoning.

When to use: Considering a $500-5000 course, or deciding between two certifications.

Mode 12: Project — Portfolio Project Assessment

Trigger: /career-ops project <repo-url-or-description>

What it does: Evaluates a portfolio project’s impact on your candidacy:

• Code quality assessment (from repo if URL provided)
• Storytelling assessment (can this be turned into a STAR story?)
• Market signal (does this demonstrate skills in demand?)
• Improvement recommendations (what to add to make it more impressive)
• Recommended highlights for CV bullet points

Output: Project impact rating + concrete improvement plan.

When to use: Before adding a project to your CV/portfolio, or deciding whether to invest more time polishing an existing project.

The A-F Evaluation Framework (Used by Auto-pipeline & Batch)

10 weighted dimensions, scored A-F:

Composite score 0-5. Below 4.0: system recommends NOT pursuing the role.

Mode Chaining: Typical Workflows

Weekly Active Search Pattern

Monday:    scan → 30 opportunities surfaced
Tuesday:   batch → top 15 scored, 5 above 4.0
Wednesday: deep → company brief for top 5
Thursday:  pdf + apply → applications submitted
Friday:    tracker → status check, stale follow-ups

Single Opportunity Pattern

Step 1: auto-pipeline → score + PDF + tracker
Step 2: deep → company brief
Step 3: apply → form-fill suggestions
Step 4: contacto → outreach to referrer

Mid-Career Pivot Pattern

Step 1: training → assess 3 certifications for the new field
Step 2: project → assess your existing projects for fit
Step 3: scan with filters → opportunities in new field
Step 4: batch → score top 10

Setup Recap

git clone https://github.com/santifer/career-ops
cd career-ops
npm install
npm run doctor          # validate setup
# Create cv.md at root with your master resume
claude                  # opens Claude Code in this directory

Then prompt naturally: “Change archetypes to backend roles”, or “Add weighting for compensation”, and Claude modifies the modes/, config/, and templates/ files directly.

Strengths and Limitations

Strengths

• Modular design — each mode is a separate skill file, easy to customize
• Human-in-the-loop — no auto-submission, you control what gets sent
• Parallel batch processing — claude -p workers dramatically reduce evaluation time
• Self-improving — interview prep STAR stories accumulate over time

Limitations

• Bilingual support is uneven — most prompts are English-first; Japanese and other languages need manual translation
• 45+ portal scan is region-biased — US/EU-focused, weak for APAC job boards
• No native ATS integration — generates PDFs, doesn’t push directly to ATSes
• High cost on Batch mode — running 10+ Claude workers in parallel is token-intensive

Related Articles

• How to Use Career-Ops on Claude Code: From Zero to Production
• Inside Career-Ops: Skills Architecture Deep Dive
• Career-Ops × Claude Code: Complete Guide
• 15 Best Claude Code Skills You Should Install in 2026
• Claude Code Subagents: Best Practices

FAQ

Which Career-Ops mode should I start with?

Default. Run /career-ops and read the mode list. Then for your first real use, try Auto-pipeline on a single opportunity — it’s the fastest end-to-end demonstration of what Career-Ops does.

How is Batch different from Pipeline?

Batch runs evaluations in parallel via claude -p workers (fast for 10+ opportunities). Pipeline runs sequentially through a queue (better for ≤10 opportunities). Batch is more token-intensive but completes faster.

Does Career-Ops actually submit applications?

No. Career-Ops emphasizes human-in-the-loop. The Apply mode generates form-fill suggestions; you submit manually. This is intentional.

What’s the A-F score threshold for “should I apply”?

The default recommended threshold is 4.0/5. Below 4.0, the system flags the opportunity as low-fit. You can adjust the threshold in config/ if you want stricter or more permissive filtering.

Can I add a new mode?

Yes. Create a new file in modes/<my-mode>.md following the existing pattern. The mode is auto-discovered when Claude Code reads the directory.

Does Career-Ops integrate with ATSes like Greenhouse or Lever?

Not directly. Career-Ops generates PDF CVs that you upload to ATS manually. There’s no API push.

What’s the token cost of Batch mode?

Each worker uses ~10K-30K tokens depending on the depth of the JD and your cv.md. For 10 workers, expect 100K-300K total tokens per Batch run. Plan accordingly.

Is Career-Ops free?

The repository itself is open-source on GitHub (MIT license). You pay for the Claude API tokens / Claude subscription used during evaluation.