I built my job hunt as a set of Claude Code skills inside an Obsidian vault

#ai #productivity #opensource #career

I'm job hunting. Rather than living in a dozen browser tabs and a stack of half-edited CV versions, I built the process into my Obsidian vault as a set of Claude Code skills (job-hunt-*) and open-sourced it as part of my second-brain template.

This is a walkthrough of how it's wired — steal whatever's useful.

The core idea: two spines, everything else orbits them

The whole toolset hangs off two plain-Markdown files:

File	Built by	Read by
`cv.md`	`cv-from-doc` (from my real CV PDF)	portfolio, mock interviews, company scouting, career-ops
`my-stories.md`	`capture-story` (STAR, one story at a time)	interview prep, mock interviews, company scouting

Because cv.md is symlinked into career-ops, there's one spine, not two copies. Fix a metric once and my portfolio, my role scoring, and my "me vs the JD" matrix all improve at the same time. That single property is what makes the rest worth building.

The skills

craft-profile — turns cv.md into a distinctive single-page HTML portfolio. It's not a CV writer; it interrogates each line until it actually understands the impact, then renders it.
capture-story / interview-prep / mock-interview — grow the STAR bank, draft story-backed answers (and flag the questions I have no story for instead of fabricating one), then drill me cold and score each answer for STAR completeness and red flags.
scout-company — complements automated role research with the Glassdoor/Blind employee-voice angle and a head-to-head matrix against the job description, rendered as a shareable dossier.
interview-debrief — a post-interview retro that mines new stories back into the bank.

The engine I didn't build

Finding and scoring roles is a big job, so I didn't reinvent it — the toolset adopts career-ops (open source, by Santiago Fernández), which scans boards and deep-scores each role A–F against cv.md. Two of my vault skills (careerops-profile, careerops-portals) exist purely to translate my CV into career-ops' config so the scan is signal, not noise. The vault makes me sharp; career-ops finds the targets.

The flywheel

my-stories.md → interview-prep + mock-interview → real interview
my-stories.md → scout-company                   → real interview
real interview → interview-debrief → (capture-story) → my-stories.md

Every interview feeds the bank, which makes the next prep, drill, and dossier better. A weak answer a mock exposes becomes a prompt to capture the missing story.

Two non-negotiables

Nothing is auto-sent. career-ops never submits an application or message on its own — I approve every PDF, letter, and outreach message.
Personal data stays private. The CV, stories, and dossiers are excluded from the public template by the repo's sync tooling, so the skills publish without the data they produce.