job_ops-mcp 0.3.0

package/modes/career_packet.md

@@ -0,0 +1,91 @@
+# Career Packet
+
+This file is the **active career packet** — the superset of every claim the candidate is
+allowed to make. `generate_materials(job_id)` picks subsets from this per JD; it never
+invents claims outside this set.
+
+When you run `npx job_ops-mcp init` the server seeds this from your `cv.md` +
+`config/profile.yml` and stores versioned copies in the `career_packet` table. Re-edit
+`cv.md`, then call the `update_career_packet` MCP tool to bump a new version.
+
+This file is also exposed as the `mcp-jsa://career_packet/active` MCP resource so the
+chat can reason against it directly without round-tripping through the DB.
+
+> **Note:** the headings below are a **template**. After `init` the server replaces
+> Section 1 (identity) from `config/profile.yml` and leaves the rest of the file as
+> editable scaffold. Replace `<TODO>` markers with bullets pulled from your own CV — keep
+> them concrete (verbs + metrics) and never list anything you can't defend in an interview.
+
+---
+
+## 1. Identity
+
+(seeded from `config/profile.yml` → `candidate` block on server start)
+
+## 2. Tagline alternatives
+
+The job rater picks ONE based on detected `role_category`. Write 4–6 variants — one per
+role archetype you target. Example shape:
+
+- **A. Default generalist** — "<one-line positioning that works across all your target roles>"
+- **B. ML / Applied AI Engineer** — "<role-specific tagline emphasizing your AI/ML credibility marker>"
+- **C. Forward Deployed / Solutions** — "<tagline emphasizing customer-facing delivery>"
+- **D. Builder PM / Technical PM** — "<tagline emphasizing PM scope + technical depth>"
+- **E. Data / Analytics Engineer** — "<tagline emphasizing data/infra ownership>"
+- **F. Generalist SWE / lean teams** — "<tagline emphasizing range and shipping speed>"
+
+## 3. Most recent role — bullet bank (5–8 to pick from)
+
+Source: `cv.md` work-experience section. One sentence per bullet, action-verb start,
+~15–25 words, real metrics from your CV.
+
+- <TODO bullet 1 — what you owned, the metric you moved>
+- <TODO bullet 2>
+- <TODO bullet 3>
+- <TODO bullet 4>
+- <TODO bullet 5>
+
+## 4. Previous role — bullet bank (3–5 to pick from)
+
+- <TODO bullet 1>
+- <TODO bullet 2>
+- <TODO bullet 3>
+
+## 5. Earlier role(s) — bullet bank (2–4 to pick from)
+
+- <TODO bullet 1>
+- <TODO bullet 2>
+
+> If you have more than three jobs worth highlighting, add new sections (`## 6. ...`) in
+> the same shape. `generate_materials` will pick subsets per JD; only the bullets in this
+> packet are eligible for the tailored resume.
+
+## 6. Projects bank (pick 2–3 per resume)
+
+- **<Project name>** — <one-sentence description with the credibility marker and stack>.
+  Link or GitHub if public.
+- **<Project name>** — <description>.
+- **<Project name>** — <description>.
+
+## 7. Skills bank (categorized — reorder per JD)
+
+- **AI / LLM Systems:** <list the tools / frameworks / paradigms you can defend>
+- **Data & Analytics Engineering:** <SQL flavours, warehouses, ETL, libs>
+- **Infrastructure & DevOps:** <containers, clouds, networks>
+- **Product:** <PRDs, frameworks, processes>
+- **Web & Tools:** <stacks, no-code, design>
+- **Languages:** <primary first>
+
+## 8. Education
+
+- **<Degree>** — <Institution> (<year range>). <optional 1-line context>
+
+## 9. Hard rules
+
+- **Never invent metrics.** Only numbers that already appear here or in `cv.md` are usable.
+- **Never surface visa / work-auth** in any resume bullet, cover letter, or outreach DM.
+  Visa data is internal scoring only — and the whole visa surface can be disabled via
+  `MCP_JSA_VISA_SCORING=false` if it doesn't apply to you.
+- **Never use cliché phrases** ("passionate about", "leveraged", "spearheaded",
+  "facilitated", "synergies", "robust", "seamless", "cutting-edge", "innovative",
+  "results-oriented", "proven track record").

package/modes/negotiation_playbook.md

@@ -0,0 +1,64 @@
+# Negotiation Playbook
+
+Loaded as an MCP resource. Used by `negotiation_brief(job_id)` together with the comp
+`enrichment` row for that company.
+
+## Frame
+
+You are negotiating from a position of evidence, not need. Three pillars:
+
+1. **Salary framework** — anchor on total comp (base + sign-on + equity + bonus + benefits),
+   not base alone. Levels.fyi / Glassdoor numbers + the offer in hand define the band you
+   work within.
+2. **Geographic-discount pushback** — refuse silent geo-adjusted offers when the work is
+   identical to a higher-band geo. Ask explicitly: "is the band the same for <your city>
+   remote as it is for the SF office?" If different, ask what specifically changes about
+   the role.
+3. **Competing-offer leverage** — never lie about an offer that doesn't exist. Real offers
+   in hand, or genuine recruiter conversations with named companies + stages, are the
+   only valid forms.
+
+## Scripts (adapt — do not paste verbatim)
+
+### Initial response to first offer
+> "Thanks — really excited about the role. Before I commit, can we walk through total comp
+> together so I understand the full picture (base, sign-on, equity vest, refresher policy,
+> bonus target, benefits)? Once I have that I can give a clean answer."
+
+### Counter — base
+> "Based on Levels.fyi for [role][level] at companies in [comparable group], the band I'm
+> seeing is [X–Y]. Given my <strongest differentiator from career_packet>, I'd land at
+> [Y]. Can you check if there's room there?"
+
+### Counter — equity
+> "Can you help me understand the equity grant — preferred-price strike, last 409a, refresh
+> schedule? At early stage I weight equity heavily, so the actual delta in [grant value]
+> vs [target] matters more to me than base."
+
+### Pushback — geographic discount
+> "I noticed the band looks different by location. Help me understand: is the *role* the
+> same in <your city> as it is in SF? Same scope, same expectations? If yes, I'd want to be
+> in the same band — comp tied to the role rather than the zip code."
+
+### When they ask for a number first
+> "I'd rather hear yours — you have more data on the band than I do. I can share what I'm
+> seeing in market data once I see the structure of the offer."
+
+## Knobs to push beyond base
+
+- Sign-on bonus (one-time, easiest yes)
+- Equity refresh schedule (annual refresh > grant size for long retention)
+- Title (one level up costs the company nothing if the actual scope matches)
+- Remote allowance / coworking stipend
+- Start date (gives time to wrap commitments)
+- Review cadence (6-month review with clear promotion criteria)
+- Relocation if you'll move within first year
+
+## Hard rules
+
+- **Never accept verbally same-day.** Always ask for the offer in writing and a few days.
+- **Never lie about competing offers.** If you don't have one, don't claim one.
+- **Visa / OPT stays out of negotiation conversation** until the company has *named the
+  comp* in writing. Then it's standard FTE onboarding — treat as logistics, not leverage.
+- **Don't negotiate against yourself.** If they ask "what would it take," give one number
+  and let them respond — do not lower preemptively.

package/modes/outreach_tone.md

@@ -0,0 +1,80 @@
+# Outreach Tone
+
+Distilled from JSA study guide §4.4 (Warm Outreach) + §4.5 (Founder Outreach).
+
+## Hard rules — apply to every outreach draft
+
+- **No emojis. No exclamation marks.** No "I hope this finds you well." No "I'd love to
+  pick your brain."
+- **No mention** of work authorization, EAD, OPT, visa, sponsorship, or anything legal.
+  The play is relationship now, action later.
+- **Never ask "refer me"** or "can you put my resume in front of someone."
+- **Soft close.** End with "no worries if not a fit" (warm) or "no worries if too busy /
+  genuinely curious / happy to be ignored" (founder).
+- Lowercase casual fine. Native tech English. Short sentences, action verbs.
+
+## Warm intro DM (existing LinkedIn connection at the company)
+
+| Constraint     | Value                                                        |
+|----------------|--------------------------------------------------------------|
+| Character cap  | **< 600 chars total (hard limit)**                           |
+| Lead           | ONE specific thing about *their* work, company, or role they hold — NOT a generic compliment |
+| Ask            | Make it SMALL: a 15-minute chat, a curious question about the team, or how they got into [role] |
+| Persona        | Peer / builder, NOT job seeker — drop one tiny credibility marker from your career_packet (a side project, a shipped product, a public artifact — keep it under 10 words) |
+| Close          | "no worries if not a fit"                                    |
+
+**Variants by connection type:**
+
+- **Engineering peer** → curious technical question about the team / stack
+- **Recruiter** → slightly more direct ("curious if there is still room in [role]") but
+  still no "refer me"
+- **Leadership** → one thoughtful question about how they're thinking about
+  [team/space] — make them want to reply
+
+## Founder DM (peer-to-peer, non-stealth founder/CEO/CTO/c-suite)
+
+| Constraint     | Value                                                        |
+|----------------|--------------------------------------------------------------|
+| Character cap  | **< 300 chars total (hard limit — shorter than warm)**       |
+| Lead           | "saw you're building {resolved_company or company_raw}" + ONE specific, curious, technical question about what they're working on (architecture choice, the wedge, the GTM) |
+| Bridge         | Drop ONE adjacent candidate project (1 short phrase): "I've been building X, curious how y'all approach Z" |
+| Persona        | Peer with adjacent project — NOT looking for a job           |
+| Close          | "no worries if too busy" or "genuinely curious / happy to be ignored" |
+
+**Adjacent project bank** — populate from `career_packet.md` section 6. Each entry should
+be one short phrase you could drop into a 300-char DM. Example shape:
+
+1. **<Project name>** — <one-line credibility marker>. Useful when their company is
+   <domain X / Y / Z>.
+2. **<Project name>** — <one-line>. Useful when <domain>.
+3. **<Project name>** — <one-line>. Useful when <domain>.
+
+**Forbidden moves:**
+
+- NOT a job ask. NO "I'm looking for a job", NO "are you hiring", NO "refer me".
+
+## Followup DM (sent-but-unanswered after N days)
+
+1–2 line nudge. NO new ask. Tone: "still curious about X — drop me a line if it ever makes
+sense." Soft close again. No guilt.
+
+## Reply draft (someone replied — chat now drafts the human's response)
+
+Match their energy. If they replied warmly with a question, answer it tightly and add one
+forward-moving piece (link to a project, a specific time window for a chat). If they
+deflected politely, send 1-line thanks + an open door ("happy to come back if anything
+opens up — no pressure either way"). Never beg.
+
+## Output contract (chat / api)
+
+When drafting, return STRICT JSON:
+
+```json
+{
+  "message":      "the full DM, within the char cap, no emojis",
+  "opening_hook": "the specific thing you led with, 1 sentence",
+  "primary_ask":  "what you actually asked them to do, 1 short phrase",
+  "strategy_note": "1-2 sentence note on why this framing should work for this contact",
+  "subject_line":  "email subject if channel=email, else 1-line placeholder"
+}
+```

package/modes/report_format.md

@@ -0,0 +1,83 @@
+# Evaluation Report Format
+
+When `evaluate_job(input, mode="chat")` returns the rubric + normalized JD, you (the chat
+client) write the 6-block report below and POST it back. The server persists it to
+`eval_reports`, renders an HTML view, and returns a localhost link.
+
+Keep blocks tight — the report is for *you* to act on, not a deliverable.
+
+## Block A — Role Summary
+
+Single table:
+
+| Field            | Value                                                  |
+|------------------|--------------------------------------------------------|
+| Archetype        | one of the 6, or hybrid (e.g. "Agentic / FDE")         |
+| Domain           | platform / agentic / LLMOps / ML / enterprise          |
+| Function         | build / consult / manage / deploy                      |
+| Seniority        | intern → principal                                     |
+| Remote           | full / hybrid / onsite                                 |
+| Team size        | if mentioned                                           |
+| TL;DR            | 1 sentence                                             |
+
+## Block B — CV Match
+
+Table: each JD requirement → exact line(s) in `cv.md`. Then a **Gaps** subsection with one
+row per gap: hard blocker vs nice-to-have, adjacent experience the candidate has, mitigation
+phrase for the cover letter.
+
+## Block C — Level Strategy
+
+1. **Level detected in the JD** vs **candidate's natural level for that archetype**
+2. **"Sell senior without lying" plan** — concrete phrases, achievements to highlight,
+   how to frame your most senior-shaped experience (PRD ownership, team lead, end-to-end
+   delivery) as senior-equivalent scope
+3. **"If they downlevel me" plan** — accept if comp is fair, negotiate 6-month review, clear
+   promotion criteria
+
+## Block D — Comp & Demand
+
+Table with market data (Glassdoor / Levels.fyi / Blind / news) + cited URLs. Pull from
+`enrichment` table for that company when `kind = 'comp'` exists. If no data, state that
+rather than inventing numbers.
+
+## Block E — Personalization Plan
+
+Top 5 changes to CV + top 5 changes to LinkedIn to maximize match. Table:
+
+| # | Section | Current | Proposed | Why |
+|---|---------|---------|----------|-----|
+
+## Block F — Interview Plan
+
+6–10 STAR + Reflection stories mapped to JD requirements. Table:
+
+| # | JD requirement | Story title | S | T | A | R | Reflection |
+|---|----------------|-------------|---|---|---|---|------------|
+
+Append new stories to `story_bank` via `extract_stories(job_id)`. Over time the bank holds
+5–10 master stories you adapt per interview.
+
+Also include:
+- 1 case study (which project to present, framed for this role)
+- Red-flag questions and how to answer them ("why did you sell your company?",
+  "do you have direct reports?")
+
+## Block G — Posting Legitimacy (optional, career-ops parity)
+
+Three tiers: **High Confidence** / **Proceed with Caution** / **Suspicious**. Signals:
+
+- Posting age (under 30d good, 30–60 mixed, 60+ concerning — adjusted for role type)
+- Apply button active
+- Tech specificity vs boilerplate ratio
+- Internal contradictions (entry-level title + staff requirements)
+- Recent layoff news
+- Reposting pattern (same role 2+ times in 90 days)
+- Salary transparency (low-reliability signal)
+
+Present observations, not accusations. Every signal has legitimate explanations.
+
+## Keywords
+
+15–20 keywords from the JD for ATS optimization — verbatim phrases the future tailored
+resume should preserve.

package/modes/rubric.md

@@ -0,0 +1,119 @@
+# Rating Rubric
+
+Distilled from career-ops' scoring system + the JSA-style three-dimension formula. Edit
+this file to retune scoring; the MCP server loads it at runtime and serves it as a
+resource to the chat.
+
+> **Visa scoring is optional.** Set `MCP_JSA_VISA_SCORING=false` to drop `visa_fit` from
+> the formula and switch to renormalized weights (`resume 0.6 + taste 0.4`). When disabled,
+> the server prepends an override block to the top of this rubric, hides the visa-related
+> tools (`visa_signal`, `import_h1b`, `import_linkedin`), and strips visa columns from
+> tool responses. If sponsorship is irrelevant to you (US citizen, non-US user, etc.),
+> turn it off and use the 2-dimension form.
+
+## About the candidate
+
+The candidate's identity, experience summary, target roles, and constraints live in
+`config/profile.yml` (seeded by `init`) and the active `career_packet`. **Read those
+before scoring** — every dimension is judged against the candidate's actual background,
+not a hypothetical average applicant.
+
+## Role priority order
+
+Defined in `config/profile.yml` → `target_roles.archetypes`. Each archetype has a `fit`
+band:
+
+- **primary** — dream role; you'd take it tomorrow at the right comp
+- **secondary** — good fit; you'd take it if the company is right
+- **adjacent** — stretch; you'd consider it if the rest is excellent
+
+If a job clearly maps to one of the listed archetypes, set `role_category` accordingly:
+`pm | ml_eng | data_eng | analytics_eng | swe | forward_deployed | other`.
+
+**Archetype override:** if the user has set `declared_archetype` on a job, use that
+instead of inferring `role_category` — it represents an explicit preference and wins.
+
+## Rating dimensions (each 0–100)
+
+### `resume_fit` — how well does the candidate's background match the role's stated requirements?
+
+- 90+ excellent match
+- 70–89 strong match with some gaps
+- 50–69 stretch but not absurd
+- <50 role wants very different background
+
+### `taste_fit` — does this role match what excites the candidate?
+
+Compare the JD against `config/profile.yml` → `narrative.likes` and
+`narrative.dislikes`, plus the active career_packet's positioning.
+
+- 90+ role + company strongly match preferences
+- 70–89 good match
+- 50–69 neutral
+- <50 conflicts with stated dislikes
+
+### `visa_fit` — will this company / role work for someone who needs visa sponsorship?
+
+Only applied when `MCP_JSA_VISA_SCORING=true`. Pull from:
+
+- The job description itself (mentions of sponsorship, work authorization, citizenship)
+- The company's H1B record via `visa_signal(company)` (requires `import_h1b` to have been
+  run with a DOL OFLC CSV)
+- The candidate's situation in `config/profile.yml` → `location.visa_status`
+
+Scoring:
+
+- 90+ company actively sponsors, role is standard FTE
+- 70–89 company sponsors but role unclear
+- 50–69 unknown sponsorship, small company, no clear signal
+- <50 contract role / US-citizens only / no-sponsor / region-locked away from candidate
+
+### `score_total` — weighted
+
+When visa scoring is **on**:
+
+```
+round( 0.5 * resume_fit + 0.3 * taste_fit + 0.2 * visa_fit )
+```
+
+When visa scoring is **off** (the server enforces this server-side too, regardless of
+what the chat returns):
+
+```
+round( 0.6 * resume_fit + 0.4 * taste_fit )
+```
+
+Tier shorthand: A ≥ 85, B 75–84, C 60–74, D 40–59, F < 40.
+
+## Output contract (chat mode)
+
+When the chat client returns scores via `evaluate_job` it MUST emit STRICT JSON:
+
+```json
+{
+  "resume_fit": 0,
+  "taste_fit": 0,
+  "visa_fit": 0,
+  "score_total": 0,
+  "reasoning":   "2–3 sentences on main fit signal",
+  "concerns":    "1–2 sentences on biggest concerns, or null",
+  "role_category": "pm | ml_eng | data_eng | analytics_eng | swe | forward_deployed | other",
+  "seniority":     "intern | junior | mid | senior | staff | principal | lead | unclear"
+}
+```
+
+(When `MCP_JSA_VISA_SCORING=false`, omit `visa_fit` — see the override block the server
+prepends.)
+
+If you cannot parse or anything is uncertain, leave `concerns` populated and set
+`role_category: "other"` / `seniority: "unclear"` — never silent zeros.
+
+## Hard rules (NEVER violate)
+
+1. **Never surface visa / work-auth in any resume, cover letter, or outreach.** Visa data
+   is internal scoring (`visa_fit`) and `visa_signal` only.
+2. **Never invent claims** not present in the career_packet / cv.md.
+3. **Human-in-the-loop everywhere** — no tool auto-submits an application or auto-sends a
+   DM. `apply_prefill` is preview-only.
+4. **Strict-JSON parsing on the api path.** On parse failure, record a `PARSE_ERROR` in
+   `score_detail` — never silently default to zeros.

package/modes/tailoring_rules.md

@@ -0,0 +1,102 @@
+# Tailoring Rules
+
+Distilled from career-ops + JSA materials generation. Used by
+`generate_materials(job_id)` to pick bullets/tagline/projects from the active career
+packet per JD.
+
+## Input
+
+You receive:
+1. The full **career packet** (superset of every claim the candidate is allowed to make)
+2. The **JD** (title, description, requirements)
+3. The detected **role_category** (or `declared_archetype` if the user set one)
+4. Optional **enrichment summaries** (comp, culture, recent_news)
+
+## Output contract (STRICT JSON)
+
+`experience_bullets` is keyed by a slug per employer / role from your career packet
+(e.g. `current_role`, `previous_role`, or whatever short employer slugs you used in
+your own packet). Use the same keys your packet uses; the renderer reads them by name.
+
+```json
+{
+  "tagline":            "from career packet section 2, pick the most appropriate alternative",
+  "experience_bullets": {
+    "<employer_slug_a>": ["\\resumeItem-wrapped bullet", "...5-8 bullets..."],
+    "<employer_slug_b>": ["...3-5 bullets..."],
+    "<employer_slug_c>": ["...2-4 bullets..."]
+  },
+  "projects_section": "full LaTeX string with 2-3 project headings",
+  "skills_section":   "LaTeX string with reordered \\item categories",
+  "cover_letter_body": "250-350 words plain prose, NO latex",
+  "tailoring_notes":  "why you picked these specific bullets/projects"
+}
+```
+
+## Tailoring decisions by role type
+
+Pick the tagline + the experience emphasis + the project order + the skills order based
+on `role_category`. The exact lead bullets depend on what's in *your* packet — the
+guidance below is about *which dimensions to surface*, not specific projects.
+
+| `role_category`              | Tagline option | Lead with                                          | Skills order                                       |
+|------------------------------|----------------|----------------------------------------------------|----------------------------------------------------|
+| `ml_eng` / Applied AI        | B              | AI / agent / model work; tool-calling pipelines    | AI/LLM → Data Science → Stacks → Product → Infra   |
+| `forward_deployed`           | C              | Customer-facing delivery; shipped enterprise wins  | AI/LLM → Product → Stacks → Infra                  |
+| `pm`                         | D              | Product ownership; PRDs; cross-functional leadership| Product → AI/LLM → Data Science → Infra            |
+| `data_eng` / `analytics_eng` | E              | Pipeline + warehouse + observability work          | Data Science → Infra → AI/LLM → Stacks             |
+| `swe` (generalist, lean)     | default A or F | Full-stack shipping evidence                       | Stacks → AI/LLM → Infra → Data Science             |
+| `other`                      | default        | Mirror highest-fit role from the table             | Product → AI/LLM → Data Science                    |
+
+## Bullet formatting (CRITICAL)
+
+- Wrap every bullet exactly: `\resumeItem{...}` (in JSON: `\\resumeItem{...}`)
+- Bold 1–3 things per bullet with `\textbf{...}` — tools, metrics, key claims
+- Escape LaTeX specials inside bullet text: `&` → `\&`, `%` → `\%`, `$` → `\$`,
+  `#` → `\#`, `_` → `\_`
+- One sentence, ~15–25 words, action-verb start
+- **Use only metrics that appear in the career packet / cv.md.** DO NOT invent numbers.
+
+## Projects section
+
+Build a complete LaTeX `\resumeProjectHeading` block. 2–3 projects total, 2–4 bullets each.
+Format:
+
+```latex
+\resumeProjectHeading
+  {\textbf{<project name>} --- <short description> $|$ \href{<url>}{<display>}}{<years>}
+  \resumeItemListStart
+    \resumeItem{<bullet emphasising the credibility marker for this project>}
+    \resumeItem{...}
+  \resumeItemListEnd
+```
+
+## Skills section
+
+```latex
+\item \textbf{AI / LLM Systems:} <comma-separated list ordered by JD relevance>
+\item \textbf{Data Science & Engg.:} <list>
+\item \textbf{Product:} <list>
+\item \textbf{Infra & DevOps:} <list>
+\item \textbf{Stacks & Tech:} <list>
+```
+
+Pick which categories to include + their order based on JD relevance.
+
+## Cover letter rules
+
+- 250–350 words, plain prose, **NO LaTeX commands at all**
+- Open with ONE specific observation about the company / product pulled from the JD
+- Mid: connect 1–2 candidate projects/experiences to the role's needs
+- Close: state interest as a conversation — substantive, not "looking forward"
+- No "I am writing to apply for ..."
+- No exclamation points, no emojis
+- **DO NOT mention work authorization / EAD / visa / sponsorship**
+- Optional: one unexpected detail per cover letter (a side project, a non-obvious angle on
+  the role)
+
+## Tailoring notes
+
+In `tailoring_notes`, explain in 2–3 sentences:
+- Why you picked these specific bullets/projects (which JD signals they match)
+- What you deliberately left out and why

package/package.json

@@ -0,0 +1,67 @@
+{
+  "name": "job_ops-mcp",
+  "version": "0.3.0",
+  "description": "Self-hosted MCP server for the full job-search loop: portal scanning, JD evaluation, tailored resume + cover PDFs, outreach drafting, story bank, negotiation brief — chat-driven, human-in-the-loop.",
+  "type": "module",
+  "private": false,
+  "license": "MIT",
+  "homepage": "https://github.com/mohith-das/job_ops-mcp#readme",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/mohith-das/job_ops-mcp.git"
+  },
+  "bugs": {
+    "url": "https://github.com/mohith-das/job_ops-mcp/issues"
+  },
+  "keywords": [
+    "mcp", "model-context-protocol", "claude", "job-search", "resume",
+    "cover-letter", "ats", "career", "agent", "playwright"
+  ],
+  "bin": {
+    "job_ops-mcp": "dist/cli.js",
+    "job-ops-mcp": "dist/cli.js"
+  },
+  "files": [
+    "dist/",
+    "modes/",
+    "templates/",
+    "fonts/",
+    "cv.example.md",
+    "config/profile.example.yml",
+    "portals.example.yml",
+    ".env.example",
+    "README.md",
+    "LICENSE"
+  ],
+  "scripts": {
+    "build": "tsc && npm run copy-assets",
+    "copy-assets": "node -e \"const{cpSync,existsSync,mkdirSync}=require('fs');mkdirSync('dist/migrations',{recursive:true});cpSync('src/migrations','dist/migrations',{recursive:true});\"",
+    "start": "node dist/cli.js start",
+    "init": "node dist/cli.js init",
+    "doctor": "node dist/cli.js doctor",
+    "connect": "node dist/cli.js connect",
+    "dev": "tsx watch src/cli.ts start",
+    "typecheck": "tsc --noEmit",
+    "prepublishOnly": "npm run build",
+    "playwright:install": "playwright install chromium"
+  },
+  "dependencies": {
+    "@modelcontextprotocol/sdk": "^1.0.4",
+    "better-sqlite3": "^11.3.0",
+    "express": "^4.21.0",
+    "js-yaml": "^4.1.0",
+    "playwright": "^1.48.0",
+    "zod": "^3.23.8"
+  },
+  "devDependencies": {
+    "@types/better-sqlite3": "^7.6.11",
+    "@types/express": "^4.17.21",
+    "@types/js-yaml": "^4.0.9",
+    "@types/node": "^22.7.4",
+    "tsx": "^4.19.1",
+    "typescript": "^5.6.2"
+  },
+  "engines": {
+    "node": ">=20"
+  }
+}