/tailor-resume

name: tailor-resume
description: Choose the best existing resume base/variant for a job, or create a new tailored variant when nothing fits.
argument-hint: "<digest-json | job-url | pasted-jd-text> [--base <resumeId>]"

Tailor Resume - Reuse or Create

Choose or produce a resume for a specific job. You decide reuse vs create; the user does not pre-select.

Setup

Follow `../_shared/setup.md`. The profile response includes `resumes` (every base with `label`, `hasData`, `variantCount`, `isPrimary`).

Step 1: Build the JD object

Detect the argument shape:

• Starts with `{` → parse as digest JSON. **No navigation, no snapshot.**
• Starts with `http` → `browser_navigate`, then `browser_snapshot` the posting body (per `../_shared/browser-tips.md`) and build the digest (`../_shared/digest-schema.md`) from it.
• Otherwise → pasted JD text; parse the same fields manually.

From the digest (`title`, `requirements[]`, `responsibilities[]`, `skills[]`, `yearsExperience`, `descriptionExcerpt`), assemble:

• `title`, `domain` (fintech/healthtech/devtools/…), `standouts` (clearance, on-call, on-site, …).
• `roleFamily` ∈ `frontend | backend | fullstack | mobile | data | ml | devops | qa | other` - match `title` + `descriptionExcerpt` against: frontend (`frontend`, `ui`, `react`, `vue`, `angular`), backend (`backend`, `api`, `services`), fullstack (`full-stack`), mobile (`ios`, `android`, `react native`, `flutter`), data (`data engineer/scientist`, `analytics`, `etl`), ml (`ml`, `ai engineer`, `mlops`), devops (`devops`, `sre`, `platform`, `infrastructure`), qa (`qa`, `sdet`, `test engineer`).
• `seniority` ∈ `junior | mid | senior | staff | lead` - from title (`junior`/`entry` → junior; `senior`/`sr.` → senior; `staff` → staff; `lead`/`principal` → lead; else mid). Cross-check `yearsExperience`: 0-2 junior, 3-5 mid, 6-9 senior, 10+ staff/lead.
• `keywords` - top 10 required skill terms from `skills` ∪ extracted from `requirements`. Lowercase, deduped, must-have ranked above nice-to-have.
• `responsibilityTerms` - top 5 verbs/nouns from `responsibilities` (`design`, `mentor`, `migrate`, `on-call`, …).

Step 2: Pick the Base

**Campaign choice wins.** If `--base <resumeId>` was passed (the campaign's selected resume) and that resume has `hasData` or a `sourceFilename`, use it as `BASE_ID` (skip scoring). Else **primary wins**: if `primaryResumeId` is set and that resume has `hasData` or a `sourceFilename`, use it as `BASE_ID` (skip scoring; Step 3 extracts content if missing). Otherwise score each `resumes` entry (max 10):

| Signal | Points | Rule | | ------------------- | ------ | ----------------------------------------------------------------------------------- | | Exact role-family | +4 | `label` maps to `JD.roleFamily`. | | Adjacent family | +2 | frontend↔fullstack, backend↔fullstack, ml↔data, devops↔backend. Not both. | | `hasData: true` | +1 | Enables content scoring; cheaper to tailor. | | `isPrimary: true` | +1 | | | JD keyword coverage | +0..+3 | If `hasData`, fetch base; `round(3 × matched/10)` over skills + projects + summary. | | Recency | +1 | `updatedAt` within 90 days. |

Highest wins. Tie-break: primary → most recent → lowest id. If no candidate has `hasData` AND no `sourceFilename`, stop:

> No usable base resume. Upload a PDF at <$JOBPILOT_WEB/resumes>, or fill a resume's editor manually, then re-run.

Let `BASE_ID` be the chosen id.

Step 3: Extract Structure if Missing

curl -fsS -H "authorization: Bearer $JOBPILOT_API_TOKEN" "$JOBPILOT_API/api/resumes/$BASE_ID"

If `content` is `null`, delegate to extract-resume so the logic stays in one place:

> Run the `extract-resume` skill for `$BASE_ID` and wait for it to finish.

Refetch the base row afterward - Step 5 needs the saved `content`. If extract-resume stops because there's no `sourceFilename`, surface the same message and stop.

Skip this step when `hasData: true`.

**Check `profileMismatches` on the base response.** Non-empty means the recruiter reads one address and the form submits another. Echo once, don't block the apply:

> ⚠ resume disagrees with your profile: {field} says "{resume}", profile says "{profile}". Fix at $JOBPILOT_WEB/resumes/{baseId}

Step 4: Decide Reuse vs Create

curl -fsS -H "authorization: Bearer $JOBPILOT_API_TOKEN" "$JOBPILOT_API/api/resumes/$BASE_ID/variants"

**Shortlist first.** Rank the list response by title similarity and fetch `GET /api/resumes/variants/<id>` for the **top 5** only - a base with 60 variants would otherwise cost 60 fetches per job.

Compute `reuseScore` (0-100) for the shortlist. Variants failing the role-family gate (different family AND not adjacent) score 0. Skip `Suggested rewrite` variants - they are not tailored for any job.

| Component | Max | Calculation | | -------------------- | --- | ---------------------------------------------------------------------------------------------------- | | Keyword coverage | 40 | `40 × matched/10` of `JD.keywords` across skills + project keywords + summary + bullets. | | Title similarity | 20 | `20 ×` Jaccard token overlap of `JD.title` vs `variant.label`, stripping `engineer/senior/the/at/-`. | | Responsibility cover | 15 | `15 × matched/5` of `JD.responsibilityTerms` in summary + bullets. | | Seniority alignment | 15 | Exact 15; one step off (mid↔senior, senior↔staff) 8; further 0. | | Domain match | 5 | `JD.domai