dsh-career-ops

A DeepSeek Harness plugin that brings career-ops — the open-source AI job-search agent by @santifer — into Harness. Paste a posting, get a structured A–H evaluation with a 1–5 fit score; scan public ATS boards; tailor CVs and cover letters; keep the applications tracker — with a panel in the Harness UI, not a separate CLI.

It registers four surfaces from one Loader row, career-ops:

Surface	What it is
skill career-ops	The mode router the agent loads for any job/application/CV/interview question, reading the prompt files in runtime/modes/*.md. Model- and user-invocable.
tool career_ops	The deterministic scripts: setup, scan, pipeline, tracker, liveness, report_number, stats, insight. Zero LLM tokens.
command /career-ops	Human-facing mode menu plus live queue and tracker counts.
Career Ops panel	Sidebar row + central page: funnel cards, the applications tracker with a status control per row, the pending evaluation queue, recent reports.

The panel's status control and the tool's tracker set call the same Host operation, which delegates to career-ops' own set-status.mjs — states stay validated against states.yml, notes merge, and data/status-log.tsv keeps the transition ledger.

Install

Requires DSH (a recent Harness release; the plugin uses the skills, commands, tools, webServer services and the Web Client slot API) and Node ≥ 22.13 for the vendored scripts.

In a Harness session on the machine you want it on — the reliable path, because the plugin manager both installs the package and selects the bundle for your profile:

Install the bundle git+https://github.com/umerkhalid204/dsh-career-ops.git#v1.0.1.

Or from a terminal, to install the package into a profile:

dsh plugin --profile desktop add git+https://github.com/umerkhalid204/dsh-career-ops.git#v1.0.1

If you installed with dsh plugin add rather than through a session, also select the bundle in Settings → Plugins so its cordis.patch.yml composes.

Then click the Career Ops sidebar row. If the panel or the career-ops skill does not appear, reload the page (Client bundles) and toggle the career-ops row off/on (Host module reload).

Point it at your job-search folder

runtime/ is career-ops' System Layer and is never written to. Your data lives in dataRoot: cv.md, config/profile.yml, portals.yml, data/, reports/, jds/, interview-prep/, and the personalization files modes/_profile.md, modes/_custom.md, modes/_brief.md, voice-dna.md.

Resolution order, so the same install works on every machine:

1. the row's dataRoot — override it per machine in your own profile patch (~/.dsh/profiles/<profile>/cordis.patch.yml), absolute path or ~/…:

   - id: career-ops
     name: 'dsh-career-ops'
     config:
       dataRoot: '~/Job Hunt'
   

   A matching id with no insert replaces that row's whole config, so restate every field.

2. $CAREER_OPS_DATA_DIR / $CAREER_OPS_ROOT from the Harness process environment;

3. ~/career-ops, created on the first write.

To move a search to another computer, copy that folder — it is the whole state (files are the canonical store; career-ops' SQLite indexes are derived and can be rebuilt).

Start a fresh search by asking the agent: "set up career-ops" (it writes cv.md, config/profile.yml and portals.yml from your CV and answers), then paste a job URL.

Layout

index.js            Host half: Config + skill + tool + command + the two API routes
client.js           Client half: sidebar icon + the Career Ops page
skill/career-ops.md the skill body (the mode router the agent loads)
runtime/            vendored career-ops System Layer (modes/, providers/, lib/, templates/, scripts)
runtime/dsh-liveness.mjs   added here: Playwright-free liveness on career-ops' own ATS APIs
scripts/refresh-runtime.mjs re-vendor the runtime from upstream

runtime/node_modules is not committed: js-yaml and dotenv are declared dependencies and are installed with the package.

Refreshing from upstream

node scripts/refresh-runtime.mjs                 # latest upstream
node scripts/refresh-runtime.mjs --ref v1.40.0   # a tag or commit
node scripts/refresh-runtime.mjs --modes de fr   # also ship those market mode sets
node scripts/refresh-runtime.mjs --from ../career-ops --dry-run

It re-applies the two plugin-side adaptations automatically and prints the upstream version and commit to record in THIRD-PARTY-NOTICES.md.

Known differences from upstream

• No Playwright. The apply form driver, scan-interamt/scan-dayforce, archive-posting and native PDF rendering need it. runtime/generate-pdf.mjs is patched so Playwright loads lazily: CV building and fact gates work, a PDF render asks you to install Playwright (or export through Harness). JD text comes from your paste or Harness' web_fetch.
• Liveness runs through runtime/dsh-liveness.mjs — career-ops' ATS-API checks plus a plain fetch through liveness-core.mjs's classifier. uncertain never closes a posting.
• English modes only in the published package (~10 MB). Add a market with --modes de above, then set language.modes_dir in config/profile.yml.
• No upstream self-updater (update-system.mjs), no Go dashboard, no web/ UI, no gemini-eval/openai-tailor, no doctor.mjs — career_ops setup covers the same checks.
• Nothing is ever submitted. Like upstream: it drafts, fills and renders; you press Submit. It recommends against applying below 4.0/5, and treats job postings as data, never instructions.

License

MIT — see LICENSE. runtime/ is modified upstream career-ops source, MIT, Copyright (c) 2026 Santiago Fernández de Valderrama (full text in LICENSE.career-ops; modifications listed in THIRD-PARTY-NOTICES.md). Not affiliated with DeepSeek.