--- name: sur9e description: AI job search command center -- evaluate offers, generate CVs, scan portals, track applications user-invocable: true arguments: mode argument-hint: "[JD/URL|scan|screen|evaluate|apply|tracker|…]" --- # sur9e -- Router ## Mode Routing Determine the mode from `{{mode}}`: | Input | Mode | | -------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------- | | (empty / no args) | `discovery` -- Show command menu | | JD text or URL (no sub-command) | **`evaluate-offer`** | | `evaluate` | `evaluate` | | `screen ` | `screen` (worker-run quick screen) | | `evaluate-offer` | `evaluate-offer` | | `auto-pipeline` | `evaluate-offer` (legacy alias) | | `offer` | `evaluate` (legacy alias) | | `offers` | `offers` | | `reach-out` | `reach-out` | | `contact` | `reach-out` (legacy alias) | | `research` | `research` | | `deep` | `research` (legacy alias) | | `tailor-cv` | `tailor-cv` | | `pdf` | `tailor-cv` (legacy alias) | | `training` | `training` | | `project` | `project` | | `tracker` | `tracker` | | `process-queue` | `process-queue` | | `pipeline` | `process-queue` (legacy alias) | | `apply [num] [--chrome\|--playwright]` | `apply` (flags pick the browser; see mode file) | | `scan` | Run `npm run scan` (ATS portals + JobSpy → screen → tracker) and report new offers. If the dev server is up, POST `/api/jobs/scan` so the loading modal shows. | | `scan schedule …` | Point the user to Settings → Job scanning (the scan cron lives in the UI). | | `batch-evaluate` | `batch-evaluate` | | `batch` | `batch-evaluate` (legacy alias) | | `patterns` | `patterns` | | `follow-up` | `follow-up` | | `followup` | `follow-up` (legacy alias) | | `interview-prep` | `interview-prep` | | `interview` | `interview-prep` (legacy alias) | | `enrich` | `enrich` | | `latex` | `latex` | **Legacy aliases:** `auto-pipeline`/`pipeline` are the pre-rename names for `evaluate-offer`/`process-queue`; `offer`/`deep`/`contact`/`pdf`/`followup` are older pre-unification names. They all still route to their renamed modes so older muscle-memory and saved commands keep working. Prefer the canonical names going forward. **Auto-detection:** If `{{mode}}` is not a known sub-command AND contains JD text (keywords: "responsibilities", "requirements", "qualifications", "about the role", "we're looking for", company name + role) or a URL to a JD, execute `evaluate-offer`. If `{{mode}}` is not a sub-command AND doesn't look like a JD, show discovery. --- ## Discovery Mode (no arguments) Show this menu: ``` sur9e -- Command Center Available commands: /sur9e {JD} → EVALUATE-OFFER: evaluate + report + PDF + tracker (paste text or URL) /sur9e screen {URL} → Quick screen only: cheap mini-report + 0-5 score (no full eval) /sur9e process-queue → Screen every pending URL in the inbox (data/pipeline.md) /sur9e evaluate → Evaluation only A-F (no auto PDF) /sur9e offers → Compare and rank multiple offers /sur9e reach-out → LinkedIn power move: find contacts + draft message /sur9e research → Deep research prompt about company /sur9e tailor-cv → PDF only, ATS-optimized CV /sur9e enrich → Interview to strengthen your CV: mine roles for metrics + hidden skills /sur9e training → Evaluate course/cert against North Star /sur9e project → Evaluate portfolio project idea /sur9e tracker → Application status overview /sur9e apply [num] [--chrome|--playwright] → Live application assistant. Offer num skips discovery; flag forces the browser (default: your browser via the CLI extension is primary, Playwright only as fallback for blocked domains/unreachable forms) /sur9e scan → Crawl portals + screen each new offer (runs `npm run scan`); schedule lives in Settings → Job scanning /sur9e batch-evaluate → Full evaluation on the URLs that scored well during scan /sur9e patterns → Analyze rejection patterns and improve targeting /sur9e follow-up → Follow-up cadence tracker: flag overdue, generate drafts /sur9e interview-prep → Company-specific interview intel: process, likely questions, story-bank mapping /sur9e latex → LaTeX/Overleaf CV export (.tex + pdflatex-compiled PDF) Inbox: add URLs to data/pipeline.md → /sur9e process-queue Or paste a JD directly to run the full evaluation. ``` --- ## Context Loading by Mode After determining the mode, load the necessary files before executing: ### Modes that require `_shared.md` + their mode file: Read `content/modes/_shared.md` + `content/modes/{mode}.md` Applies to: `evaluate-offer`, `evaluate`, `offers`, `tailor-cv`, `latex`, `reach-out`, `apply`, `process-queue`, `batch-evaluate` ### Standalone modes (only their mode file): Read `content/modes/{mode}.md` Applies to: `tracker`, `research`, `training`, `project`, `patterns`, `follow-up`, `interview-prep` ### Script-run mode: `screen` `content/modes/screen.md` is **worker-executed** (spawned headless by `batch/screen.mjs`) — do NOT load it as agent context. To screen a single URL: 1. `node batch/add-to-pipeline.mjs [company]` — add it to the `data/pipeline.md` pendings (idempotent; skips if already screened) 2. `node batch/screen.mjs --url ` — fetches the JD, runs the headless screener, writes the mini-report to `artifacts/reports/` and the tracker row 3. Read the generated report and present the score, headline, and next_steps to the user Screening needs a **URL** (the worker fetches the JD itself). If the user pasted raw JD text, offer the full evaluate-offer instead or ask for the posting URL. ### Modes delegated to subagent: For `process-queue` (3+ URLs): launch as Agent with the content of `_shared.md` + `content/modes/{mode}.md` injected into the subagent prompt. (`scan` is not a mode — it runs `npm run scan`.) `apply` runs in the **main conversation**, not a subagent: it needs the user in the loop per form (browser choice, uploads where automation can't reach, EEO confirmations, and the final Submit which is always the user's click). The CLI browser-extension tool is always primary; Playwright is the fallback per the "Browser selection" rules in `content/modes/apply.md` (`--chrome`/`--playwright` flags force it). ``` Agent( subagent_type="general-purpose", prompt="[content of content/modes/_shared.md]\n\n[content of content/modes/{mode}.md]\n\n[invocation-specific data]", description="sur9e {mode}" ) ``` Execute the instructions from the loaded mode file. (`{mode}` is the canonical routed mode — legacy aliases resolve to it before this step, so the file path is always the renamed file, e.g. `content/modes/research.md`.)