← ClaudeAtlas

night-shiftlisted

Use when the user wants a product continuously driven forward without them steering each step — an autonomous session (overnight or daytime) that discovers, builds, verifies, and lands improvements across every lens (user types, UX, security, compliance, reliability, DX, product strategy). Triggers on "night shift", "run the night shift", "drive this product forward", "keep improving this while I'm away", "work the queue overnight", "make this best-in-class across the board", "put the army on it". Accepts a steer argument ("night-shift focus security", "night-shift billing UX") which prioritizes that lens or area. Accepts an auto-merge argument ("night-shift auto-merge") which lands each verified improvement to main behind green checks on pre-production repos. NOT for a single decided feature (/goal-maker), a one-off QA pass (/qa), read-only scoring (coverage-matrix), or driving the coverage matrix cell-by-cell (/bhag).
jackneil/claude-jacked · ★ 1 · AI & Automation · score 67
Install: claude install-skill jackneil/claude-jacked
# Night Shift The **daily front door for autonomous product improvement**. You invoke one command; a rotating roster of **brains** (product-manager, designer, security, compliance, user personas, SRE, DX) discovers what a best-in-class product would have, files it as work, implementers build it, a skeptical evaluator proves it, and verified work lands — either as one consolidated PR for morning review (default) or merged continuously behind green checks (`auto-merge`, pre-production repos only). Every iteration ends by making the system itself smarter. The user should never have to remember sibling skills to get this outcome. Night Shift **orchestrates** them internally: `coverage-matrix` for scoring, `aesthetic-dogfood-audit` for persona crawls, `recursive-10-10-product-hardening` semantics for verification, `/bhag`'s production guard for merge safety. Those remain independently invocable for surgical use; Night Shift is the one you reach for daily. ## Core rule **Nothing lands on an agent's say-so.** Every improvement is proven before it counts: tests green, checks appear AND pass, browser evidence for UI claims, and the evaluator that grades the work is never the agent that produced it. A run that "completed 12 improvements" with 12 unverified diffs completed nothing. ## When to use - Daily driver: "keep pushing this product toward best-in-class" with no per-item steering. - Overnight/unattended runs where the morning deliverable is a reviewed-or-landed body of work