Prompt file imported from Sudeeparyan/a-resume-builder (.claude/commands/hunt.md). Fill in {{arguments}} before use. Copyright stays with the author.

/hunt — the whole loop, one command

Arguments: {{arguments}} (may be empty). Parse loosely:

• a bare number → how many applications to prepare (default 10)
• remote → Remote (US) only
• cap-exempt → universities, national labs, nonprofit research institutes, academic medical centers only
• anything else → a role/track filter (e.g. data engineer, embedded)

Run every step without asking permission in between; she asked for the whole thing. PY below means career-dashboard/backend/.venv/bin/python and WS means $PY career-dashboard/backend/scripts/workspace.py.

1. Load the facts

Read career-dashboard/AGENTS.md, then career-dashboard/data/context/ (all nine files plus QUESTIONS-FOR-YOU.md), data/config/profile.yml, portals.yml, regions.yml, sponsorship.yml, and backend/workflows/modes/_shared.md → _profile.md → batch-resumes.md. She is on active F-1 OPT: authorized to work now, sponsorship needed later.

2. Age the tracker

WS age (applications quiet 21 days become ghosted). WS summary shows what is saved, applied and excluded.

3. Find roles (~1.5× the target, the gate will cut some)

1. Tracked career pages first, no AI: WS run --kind discovery --preset portals.
2. Then AI discovery (WS run --kind discovery) and live search (Indeed search_jobs with country US, then the employer/ATS page). Always include a cap-exempt pass and sponsorship-positive queries.
3. Unless the arguments narrow it, aim for the batch mix in profile.yml: 4 Data · 2 ML/AI · 2 Software · 2 Embedded.

4. Pull the full JD and gate it

For each lead found by hand, save the full JD to a scratch file and run WS sponsor-check --company "<name>" --file <jd.txt> and WS check-reapply --company "<name>" --title "<role>". Save survivors (and exclusions, so they are logged with their sentence) with $PY career-dashboard/backend/scripts/career.py add --file <job.json> (fields: company, title, location, url, description, requisition_id). Leads found by the app were gated already.

5. Verify the link

$PY career-dashboard/.agents/skills/verify-job-url/scripts/verify_job_url.py --url "<url>". Drop dead links.

6. Rank and pick

Tier first (S → A → B → C), then score. Take the top N distinct companies. Assign N different signature projects across the batch (career-dashboard/data/signature-projects.md shows who already owns what; Pacman coursework can only support).

7. Build each application

1. $PY career-dashboard/backend/scripts/career.py prepare JOB_ID [--project PROJ-ID] → data/output/applications/Annie_Manoharan_<Company>_<NN>/.
2. Research → company-research.md (modes/deep.md). Mandatory, every time.
3. Tailor with the resume-tailor skill: registry facts only, recruiter audit, fit exactly one US Letter page (cut content in the documented order; never shrink fonts below 10pt or touch margins).
4. Validate: $PY career-dashboard/backend/scripts/validate_resume.py <folder>/resume.tex --compile --output <folder>/resume.pdf --render-dir <folder>/resume-preview --qa-json <folder>/qa.json. Look at page-01.png.
5. Study plan → study-plan.md: WS run --kind study_plan --job-id JOB_ID (or modes/upskill.md by hand).

8. Report back

data/output/SUMMARY.md regenerates itself. Print:

| # | Company | Role | Tier | Score | Location | Folder | Apply |

Then briefly: how many were found, kept and excluded, and the sentence behind each exclusion; the single skill most of these roles wanted that she does not have yet; one plain next step.

Non-negotiable

• Both resume.tex and resume.pdf in every folder.
• Every resume exactly one page, verified.
• Ten companies means ten different signature projects.
• Never surface an excluded, already-applied or recently-rejecting company.
• Never invent a listing, a link or a fact. If fewer than asked, say so and why. Never apply for her.