← ClaudeAtlas

job-underwriterlisted

Decides whether a specific job is worth applying to, and what the real gaps are. Use this whenever someone is weighing an actual role — "should I apply", "am I a fit for this JD", "is this posting worth pursuing", "is this job legit", "am I reaching", "is it worth my time" — or pastes, uploads, or links a job description with any question about fit, gaps, or whether to bother. It maps the candidate's evidence requirement by requirement, recovers truthful experience the CV left out, classifies each gap (presentation, discovery, transferability, capability, hard-gate), flags risks and red flags in the posting itself, and returns a go / conditional-go / hold / no-go decision with an application plan. Also use to vet a suspicious or vague posting. Candidate-side only — not for employers screening applicants. Interview rehearsal and prep packs are mock-interview's; when a request spans both ("should I apply, and prep me if so"), still use this skill for the decision and name the sibling for the rehearsal.
wavemanco/tuesday · ★ 0 · AI & Automation · score 72
Install: claude install-skill wavemanco/tuesday
# Job Underwriter Underwrite an opportunity on evidence, with the reasoning shown. **Evidence before language** — a polished phrase is not evidence; every material conclusion traces to a source or is labeled as inference. **Candidate fit versus opportunity quality** — the two verdicts live on separate ledgers and are never averaged together. **Started by its own command, `/job-underwriter`, this skill keeps the turn:** `references/interaction-grammar.md` §6 says what that rules out — read it before reaching for any other skill. ## Before sending Copy this list and check the response against it. Each item points to its rule and restates none; rules 1, 5 and 7 are in `references/output-discipline.md`. ```text 1. Snapshot first, read off the finished map; no recap anywhere (Step 1) 2. Every first-pass section, in Step 1's order; no questions in the gap queue (Step 1) 3. Provenance label on every derived statement (Step 1) 4. One Research tier line; one row per declared target (Step 2) 5. Questions within question_budget, counted across the message (rule 1) 6. The closing run in closing_order, nothing between its parts (Step 2) 7. Watch-out line last, exact, never in tuesday-dossier.md (rule 7) 8. Sibling named at most sibling_mentions times (rule 5) ``` ## Guardrails - **Truthfulness core.** The never-do list in `references/evidence-conventions.md` §6 binds every output of this skill; when uncertain, drop the fact rather than hedge it. Inflation hides in single words; run