Add jobs to the database (use after a web search for quant roles).

Each job dict should include at least firm and title. Useful fields: role_type (quant_dev|quant_researcher|quant_trader|quant_analyst|other), location, url, description, must_have (list), nice_to_have (list), skills (list), comp_min, comp_max, currency, posted_at. Skills are normalised, role type inferred, fit scored, and duplicates merged.

Search the job database. Empty filters are ignored. Returns job records.

Get a single job by id, including full description and skills.

Update editable fields on a job (e.g. notes, status, comp_min, role_type).

Star (save) or unstar a job you like.

Archive a job (sets status='archived'); kept for pattern history.

Set a job's status: active | applied | archived | closed.

Run a READ-ONLY SQL query against the job DB for ad-hoc pattern analysis.

Only a single SELECT/WITH statement is permitted (enforced on a read-only connection). Handy tables: jobs, job_skills(job_id, skill), skills_profile, pending_projects, skill_flags. Example: SELECT skill, COUNT(*) n FROM job_skills GROUP BY skill ORDER BY n DESC

Skill demand across tracked jobs; flags in-demand skills you don't have.

Skills you're missing (overall, for a job, or across starred jobs), each with a suggested project + hours to close the gap.

List flagged skills (recommendations to learn). Optionally filter by status.

Update a skill flag's status: new | acknowledged | learning | dismissed.

Add a project to your backlog (e.g. to earn a missing skill). Always record an honest time estimate in est_hours.

List backlog projects (optionally filter: idea|planned|in_progress|done).

Update a pending project (e.g. status, priority, est_hours, notes).

Generate a CV PDF — base (job_id=0) or tailored to a job. Tailoring only reorders/selects your real experience; missing skills become pending projects + flags rather than fabricated bullets. stretch_level: conservative | balanced | aggressive.

Start a conversational CV draft — base (job_id=0) or tailored to a job.

Returns {draft_id, content, missing_skills, suggested_projects}. content is the editable CV (contact, profile summary, education, experience-with-selected-bullets, skills_rows, volunteering, awards). Show it to the user, then refine it with cv_draft_update / cv_reframe_bullet and render with cv_draft_render.

POLICY: bullets may be reframed and strengthened (the user wants punchy bullets). Any change that goes beyond a literal restatement MUST be logged via cv_reframe_bullet (preferred — it edits and logs atomically) or cv_note_add, classified as reframe | stretch | fabrication, so it lands in cv_notes.md. Never silently invent. Missing skills are returned (and logged as pending projects).

Get the current state of a CV draft (id, content, status, timestamps).

Apply a structural edit to a CV draft (summary, ordering, skills) and return content.

Use this for the summary, reordering, and skills. For rewording individual experience bullets, prefer cv_reframe_bullet (it logs the change). Patch keys: profile (str summary), contact (dict), experience ([{org, role?, location?, start?, end?, bullets?}] merged by org), skills_rows ([{label, value}] merged by label), education/awards/volunteering (replace the list). If you reword bullets here, also call cv_note_add for anything beyond a literal restatement.

Render the agreed CV draft to a PDF. Returns {tex_path, pdf_path}.

List recent CV drafts.

Reword (or add) one experience bullet on a draft AND log it to cv_notes.md.

This is the right tool for making bullets punchy. org selects the experience entry; if original matches an existing bullet it is replaced, otherwise rewritten is added as a net-new bullet. kind MUST reflect how far you went: reframe — same facts, stronger wording (safe) stretch — plausible extrapolation/quantification the user must be ready to justify fabrication — a claim not actually done (a conscious, flagged choice) Provide justification (how the user backs it up). The change is appended to cv_notes.md.

Log a CV claim change to cv_notes.md (reframe | stretch | fabrication).

Use when you reframe/embellish a bullet or summary outside cv_reframe_bullet, so the audit trail stays complete. scope is free text (e.g. 'summary' or 'job 107').

Return the CV claim notes (cv_notes.md) — review before interviews.

Return your CV profile (contact, experience, skills, etc.) from profile.yaml.

Record that you have/are learning a skill (updates profile + fit scoring). source: have | learning | pending. Use this when you complete a project.

Run the job-board scrapers (Greenhouse/Lever). Empty = all firms; or pass a source kind ('greenhouse'/'lever') or a firm name.

Save a search/pattern so you can re-run it later.

List saved searches/patterns.

Create the database and seed your CV profile (safe to call repeatedly).

Get current settings (stretch_level, trend thresholds, location filter).

Update a setting, e.g. stretch_level=aggressive or location_filter=New York.