pm-job-search

A Claude Code plugin for senior product leaders running a job search. Finds open roles, scores fit, tracks your pipeline, runs live interview practice, and gives you an honest career coach. Markdown by default — optional browser dashboard and MCP integrations on top.

Who it's for

If you're a senior PM, lead PM, or head of product — you're in the right place. Everything inside (scoring, pipeline, interview prep, reviewer agents) is shaped around how your search actually runs. A different role? You can fork the architecture, but the defaults will need rework.

What makes it different

• Lean and fast to set up. Pure markdown, zero dependencies. One 10-15 min /setup and you're running. Everything lives in .md files under userdata/ — gitignored by default, editable by hand. Optional MCP integrations (Granola / Calendar / Gmail) plug in via /pm-job-search:integrations when you want them.
• Tailored to the senior-PM context. Reviewer agents play the four people who'll actually interview you (CPO, eng manager, design manager, interview coach), and a simulator runs mock rounds when you want to rehearse. When you're weighing an offer, it checks the role's stage against how you actually operate — builder, scaler, or operator — and flags things like hollow-HoP risk before you sign. And it pushes back when the search itself isn't working: eight weeks in with one thread, career-coach tells you adjusting cadence won't fix it.
• Visual dashboard for quick actions. /pm-job-search:dashboard opens a browser view of your pipeline — accordion grouped by status, inline status changes, notes per company, latest daily brief inline. Useful for scanning, or bumping a status after a recruiter call without opening files.

How it works

Maps to the five phases of a senior-PM search.

1. Set up once. /pm-job-search:setup — 11 questions, ~10-15 min. Writes your profile, strategy, and the rubric the rest of the plugin scores against. Drop a CV at userdata/cv.md first and setup pre-fills positioning.

2. Discover & score. /pm-job-search:job-search does a weekly sweep — parallel subagents recheck your P0/P1 companies via ATS APIs and discover new roles via scoped search. Score one-off postings ad-hoc with /pm-job-search:evaluate-position <url>. Both write the same artefacts: a company folder, a tier score, and a 200-word research brief.

3. Apply. /pm-job-search:apply <Company> tailors your master CV for one role using the research brief — three-phase process (silent analyse → upfront Q&A capped at 5 → single-pass draft). Hard rule: every claim traces to your master CV, profile, or your own answers. Cover note + outreach are user-driven follow-ups.

4. Interview. /pm-job-search:interview-prep <Company> adapts 3-5 stories from your bank for a specific round — --stage shapes the prep, late rounds add founder-vetting questions. Rehearse with pm-job-search:interviewer-simulator (mock round, single-question deep-dive, or pressure-test). Debrief with /pm-job-search:interview-analysis — quotes the transcript, flags what landed, names deltas vs your prep.

5. Decide on the offer. /pm-job-search:evaluate-offer sense-checks the offer (or compares two) against your profile, anti-goals, and the senior-PM archetype/anti-pattern references. Verdict, comp shape, archetype fit, named-anti-pattern scan, 3-5 negotiation moves. Always pairs with a "what would I regret in 2 years?" handoff to the coach.

Always running alongside:

• /pm-job-search:today — morning brief; 2-phase (input phase captures fresh facts from Calendar/Gmail/Granola, then top-3 actions + pipeline + heads-up)
• pm-job-search:career-coach — anytime you're stuck, weighing something strategic, or your search itself isn't working
• /pm-job-search:dashboard — visual pipeline view in the browser; inline status changes + notes
• /pm-job-search:story-builder — the universal STAR-story bank that /interview-prep draws from; build it up over time

Install

# inside Claude Code, in any workspace:
/plugin marketplace add https://github.com/archugunov/pm-job-search.git
/plugin install pm-job-search@pm-job-search

cd <your-workspace>
# inside Claude Code:
/pm-job-search:setup           # 10-15 min — identity, target role, salary, hard filters, target date
# you're done. run /pm-job-search:today every morning to see your daily brief.
# ask pm-job-search:career-coach anytime — stuck, got an offer, or want to sharpen your positioning / outreach.
# run /pm-job-search:help anytime for a one-page scannable map of every skill and agent (also saved to userdata/help.md).

The workflow — 13 skills

Reference:

One-time setup (~10-15 min):

Daily / weekly:

Interview cluster:

Optional — wire in external tools:

Agents — 6 personas

Four reviewers, one strategic coach, one interview-practice simulator.

The four reviewers (cpo / eng-manager / design-manager / interview-coach) share the same four-section output — what works, what doesn't, where it sounds weak from that persona's lens, one rewrite suggestion. Run 2-4 of them in parallel and read them as a panel.

Each agent has its own project-scoped memory at .claude/agent-memory/<agent>/. Reviewer agents take --save <Company> to write the review to userdata/companies/<Co>/[<role-slug>/]review-<persona>-<date>.md.

pm-job-search:career-coach is the home for any deeper strategy work. /setup writes a basic strategy.md — target date, weekly cadences derived from your timeline, headline goal — and leaves everything else (anti-goals, checkpoints, target tuning, rubric calibration) for a conversation with career-coach when you want it. Invoke by natural language — "got an offer to weigh", "I'm stuck and don't know why", "help me think through what I won't do this search". The agent figures out what's actually wrong before proposing a fix, and routes to the right mechanic (rubric tuning, anti-goal, cadence change, positioning rework — whichever it is).

Visual dashboard

A browser view of your pipeline that reads the same markdown files as every other skill — useful when you want to scan, not act.

Video walkthrough: https://github.com/user-attachments/assets/e1b39e00-d600-4ef4-bd16-ccda8025c8c4

Three panels:

• Left — Applications. Accordion grouped by status in funnel order, each header showing the count for that status. Each row has tier badge (P0/P1/P2), company + position, status dropdown for inline change, and last-touched date. Click any row to open the company drawer.
• Right top — Pipeline & weekly stats. A 4-stat bar (Offer / Active / Applied / Days Left until target offer date) plus a "This Week" panel tracking warm outreach + applications against your strategy.md targets. At-a-glance "am I tracking?" view.
• Right bottom — Today. Renders the latest userdata/outputs/daily-brief-*.md inline — top 3 actions + heads-up section — so you don't have to flip files.

Company drawer (opens on row click) has four tabs:

• Research — the research-brief.md written by /evaluate-position
• Notes — notes.md for quick thoughts between meetings; add / edit / delete in place, with conflict detection if you edit the same note from CLI elsewhere
• Prep — every interview-prep-*.md for this role, newest first
• Debriefs — every interview-debrief-*.md for this role, newest first

When to use the dashboard vs the CLI

The dashboard is intentionally read-mostly. It writes back exactly two things: status changes (to each role's meta.md) and notes (to notes.md per company). Everything else is reading. New roles, evaluations, applications, prep docs — those come from CLI skills because the model needs to be in the loop.

Start / stop

/pm-job-search:dashboard         # opens browser tab at http://localhost:<port>
# ctrl-C in the terminal to stop

React + Mantine, pre-built bundle committed to the repo. Zero install, no Node required at runtime.

Already use Notion? Skip the dashboard.

If you already run the Notion MCP, you can sync your pipeline to a Notion database instead. The markdown in userdata/companies/ stays the source of truth; Notion becomes a derived view. One on-demand sync, no auto-sync, no bidirectional drift. See INTEGRATIONS.md §4 — Notion for the prompt pattern.

Extending the plugin

Markdown-only is the default, but you can wire in MCP servers you're already running to make specific skills more capable. See INTEGRATIONS.md for the prompt patterns, covering:

• Granola — pull meeting transcripts into /interview-analysis + surface interview recordings as input-phase confirms in /today
• Gmail (or any inbox MCP) — surface recruiter replies, rejections, and offer language as input-phase confirms in /today
• Calendar (gcal / iCal) — surface upcoming interviews in /today's heads-up + flag new / rescheduled / cancelled events as input-phase confirms
• Notion — sync companies as Notion DB rows alongside the markdown source
• Playwright — link-liveness verification + interviewer research
• Slack / Telegram / Discord — pipe /today digests to an accountability channel

Each one is a prompt pattern, not a plugin code change — nothing to install on the plugin side, nothing it auto-detects.

What's in userdata/examples/

Two fictional personas so you can see what a working install looks like before running /setup:

• userdata/examples/maya/ — Maya Patel, senior PM in London doing consumer credit / fintech. Full install: profile, strategy, journal, two companies (one interviewing, one rejected), two stories with adaptation angles, a generated applications.md index.
• userdata/examples/diego/ — Diego Alvares, VP Product in Mexico City going after fully-remote US roles. Skeletal install (profile, strategy, one company, one story) — stress-tests the schema against USD compensation, no anchor city, US English, B2B SaaS.

Both personas are entirely fictional.

Privacy + contributing

Your userdata/ directory is gitignored except for the example installs. Anything you put there — profile, journal, company notes — stays on your machine unless you share it on purpose.

A CI workflow scans every push and PR for the plugin author's personal-data blocklist (via ripgrep). If you fork, swap the blocklist for your own terms or remove the workflow entirely — your fork, your call.

Contributions welcome. CONTRIBUTING.md covers scope, dev install, the privacy hard rule, testing conventions, and what's in scope for issues vs not.

Feedback & support

• Found a bug or have a feature request? Open an issue on GitHub — include the skill you ran, what you expected, and what happened. Repro steps help.

Licence

MIT. See LICENSE.