job-search

Job Search Co-Pilot

Your personal career agent with long-term memory. Handles everything from finding roles to negotiating offers, and remembers context across sessions.

Architecture

Working Directory/
├── JOBSEARCH.md          ← Profile + preferences (hot cache, loaded every session)
├── PIPELINE.md           ← Opportunity tracker (loaded every session)
├── CLAUDE.md             ← Memory hot cache (top companies, contacts, terms)
├── pipeline.html         ← Visual Kanban dashboard
└── memory/               ← Deep storage (loaded on demand)
    ├── glossary.md       ← Decoder ring: companies, contacts, acronyms
    ├── companies/        ← Cached research per company (company-slug.md)
    ├── contacts/         ← Recruiter/HM profiles + interaction history
    ├── applications/     ← Tailored materials per application
    │   └── company-role/
    │       ├── resume_tailored.docx
    │       ├── cover_letter.md
    │       ├── fit_analysis.md
    │       ├── outreach_email.md
    │       ├── interview_prep.md
    │       └── negotiation_strategy.md
    └── analytics/        ← Tracking patterns over time
        ├── applications.md
        ├── interviews.md
        ├── offers.md
        └── briefings.md

Commands

Command	What It Does
/job-search:start	Run intake interview, set up profile, pipeline, and memory
/job-search:find	Search for matching roles across the web
/job-search:research [company]	Deep-dive a target company
/job-search:apply [url]	Analyze posting, tailor resume, draft cover letter + outreach
/job-search:prep [company]	Interview prep — questions, STAR answers, research
/job-search:negotiate [company]	Salary research and negotiation scripts
/job-search:status	Pipeline dashboard — what needs attention
/job-search:briefing	Weekly digest — new matches, news, follow-ups due
/job-search:update	Sync memory, triage stale items, refresh context

See commands/ for detailed instructions per command.

Memory System

Two-Tier Architecture

Tier 1: Hot Cache (loaded every session)

Three files loaded automatically to provide instant context:

• JOBSEARCH.md — Full profile with target titles, comp floor, preferences, watchlist. The match scoring engine reads this to evaluate every job found.
• PIPELINE.md — Every opportunity tracked through 7 stages (Discovered → Offer). Drives the status dashboard and follow-up reminders.
• CLAUDE.md — Quick-reference memory (~50-80 lines): active targets, key contacts, terms decoder. Covers 90% of session lookups without hitting deep storage.

Tier 2: Deep Storage (loaded on demand)

• memory/glossary.md — Complete decoder ring. Every company, contact, acronym, and term encountered during the job search. Used when CLAUDE.md doesn't have the answer.
• memory/companies/ — Full research reports per company. Cached for 14-30 days before suggesting refresh. Each file has a timestamp header.
• memory/contacts/ — Individual profiles for recruiters, hiring managers, referrals. Includes interaction history and notes.
• memory/applications/ — All tailored materials organized by company-role. Resume, cover letter, outreach, fit analysis, interview prep, negotiation strategy.
• memory/analytics/ — Longitudinal tracking. Application counts, conversion rates, time-in-stage, offer comparisons. Powers the /briefing recommendations.

Memory Lookup Flow

1. Check CLAUDE.md hot cache (instant)
2. Check memory/glossary.md (if not found)
3. Check memory/companies/ or memory/contacts/ (for full context)
4. Web search (if nothing cached or cache is stale)
5. Ask user (last resort)

Memory Update Rules

• Always update CLAUDE.md when pipeline changes (new application, stage move, new contact)
• Always update glossary.md when encountering new companies or contacts
• Save company research to memory/companies/ after every /research command
• Save application materials to memory/applications/ after every /apply command
• Log analytics after key events (application submitted, interview completed, offer received)
• Never delete memory files — archive by prefixing with _archived_ if no longer active

Reference Guides

Detailed methodology docs loaded on demand by specific commands:

Reference	Used By	Purpose
references/intake-questions.md	/start	Conversational intake framework
references/job-discovery.md	/find	Search templates, match scoring, de-dup
references/application-toolkit.md	/apply	JD analysis, resume tailoring, outreach
references/interview-prep.md	/prep	STAR prep, question banks, day-of checklist
references/negotiation.md	/negotiate	Salary research, scripts, Indian CTC

Scripts

Automation utilities called by commands:

Script	Purpose
scripts/init.py	Initialize workspace, create files and directories
scripts/match_score.py	Score jobs against profile (0-100 with breakdown)
scripts/pipeline.py	Pipeline CRUD: status, add, move, followups, export
scripts/briefing.py	Generate weekly briefing from pipeline state
scripts/export_materials.py	Package application materials for a company
scripts/memory_sync.py	Rebuild CLAUDE.md hot cache from deep storage
scripts/analytics.py	Compute conversion rates and time-in-stage stats

Proactive Behaviors

The skill should be proactive, not just reactive:

• If user mentions a company in conversation → check if they have matching roles, check pipeline
• If user shares a job posting → immediately run gap analysis
• If an interview is approaching → remind them to prep
• If a follow-up is overdue → nudge them
• If an offer deadline is near → flag it urgently
• If user seems discouraged → show pipeline stats and progress made
• If user mentions salary/comp → check against their floor and market data
• If a new session starts → load CLAUDE.md and check for pending actions in PIPELINE.md

Session Start Behavior

At the beginning of every session where job search context is relevant:

1. Load CLAUDE.md (hot cache)
2. Quick-scan PIPELINE.md for urgent items (offers expiring, interviews this week)
3. If urgent items found, proactively mention them
4. Otherwise, wait for user to invoke a command