cv-cover-letterlisted

Draft a cover letter from a job posting and the applicant's own CV and evidence base, making only claims that trace to something they actually said, and stopping when the evidence stops rather than filling a word count. Use when the user asks for a cover letter, a covering letter, a letter of application, a motivation letter or an Anschreiben; when they want an existing draft checked for claims it cannot support; or when they ask whether a posting is worth applying to before writing one. Runs on any posting — a URL, a pasted advert, an email — not only LinkedIn. Do NOT use for the CV itself (cv-and-human) or for recovering what someone has done in the first place (cv-evidence-base).
kevin-burns/claude-skills · ★ 8 · Data & Documents · score 78

Install: claude install-skill kevin-burns/claude-skills

# Cover letter, from evidence A cover letter is a **volatile document**: a mediocre one is ignored, but a wrong one disqualifies before anything true about the applicant is read. The upside is bounded and the downside is not, so this skill biases toward fewer sourced claims over more impressive ones, and toward a shorter letter over a padded one. Its purpose is **to reduce the cost of producing a true letter**, not to produce a persuasive one. `cv-evidence-base` makes a weak CV less weak; this makes an expensive document cheap without letting it drift from the facts. ## What the evidence actually supports This skill's rules come from measured outcomes, not from career-advice convention. The full synthesis with citations is in `references/evidence.md`. The four that change behaviour: - **Composition beat tailoring.** Detail, clarity and structure predicted more interviews (β = 0.20, *p* = .016) and a shorter search; **tailoring predicted nothing** (*p* = .34). Wingate, Robie, Powell & Bourdage (2025), *Int. J. Selection and Assessment*. - **Mirroring the posting has stopped signalling.** After an AI writing tool launched on a large labour platform, the correlation between a letter's textual alignment and callbacks **fell 51%**, and employers moved to prior work history. Cui, Dias & Ye (2025). Replicated independently on different data by Galdin & Silbert (2025). - **A signal only carries information if it is costly to fake.** Spence (1973). An adjective is fre