--- name: score-fit description: Scores how well a DS/CS student's profile fits an internship, co-op, or new-grad SWE / data science / ML posting on a 1-5 scale with evidence-labelled blockers, skill gaps, and work-authorization checks; deep mode adds a why-them answer, likely interview questions, and gap fixes. Use when the user asks "should I apply", "how good a fit is this", wants roles ranked, or wants an in-depth look at one role. --- # score-fit ## Inputs - A role from `profile/roles.md` (by role_id), or a pasted JD / URL (add it via `find-roles` step 4–7 first so it gets a role_id). - `profile/facts.json`, `profile/prefs.yml`, `resources/rubric.md`. ## Steps 1. **Extract** from the JD: title, level, season/term, location + workplace type, must-haves, nice-to-haves, years required, grad window, sponsorship language, salary. Quote the JD line for each. 2. **Blockers** per rubric. Label each finding `posted` / `inferred` / `unknown`; only `posted` blockers cap the score. 3. **Score** each rubric dimension. Each match cites a fact id **and** the exact string from facts.json, e.g. `exp1: "Built Python/FastAPI service; report generation 40s -> 6s"`. A skill counts as used only if it appears in an experience or project; one that appears only in `skills` is "listed". 4. **Gaps**: missing must-haves, each with a cheap honest fix (course, small project, reframe a true fact). 5. **Verdict**: ≥4 apply now; 3.5–3.9 apply + tailor; 3–3.4 apply if low effort; <3 skip. 6. **Write** a `### Fit` block under the role in `profile/roles.md` and show: ``` cohere-36d1f5 — ML Intern/Co-op — 4.1/5 — Apply now Needs verification: co-op term "Winter 2027" vs your seasons (posted) → ask: summer start possible? Matches: Python (exp1: "Built Python/FastAPI service…"), ML (proj1: "MAE 3.1 min") Gaps: PyTorch listed only → small PyTorch project Work auth: Canada citizen ✓ (posted: "Canada") · Salary: not posted ``` 7. **Deep mode** (user asks for depth, or score ≥ 4): add a why-this-company answer (≤80 words, facts cited), 3 likely interview questions with the fact/story to use, and the 2 gaps most worth closing first. 8. Offer `track-pipeline` for scores ≥ `min_score_to_apply`, then `tailor-resume` and `find-people`. ## Rules - Inflating a score or a match is the failure this skill exists to prevent; when unsure, label `unknown`. - Market estimates are labelled `inferred` with their source and never presented as posted.