← ClaudeAtlas

darkmoonlisted

Summon an independent, grounded skeptic to challenge the author's last response. A fresh-context agent reads the response + verifies against the real code/transcript, then upholds, refutes (evidence-grade), or marks can't-determine - per claim. Grounded, never manufactured; argument-counters never masquerade as proof.
sgtlaugh/skills · ★ 2 · Code & Development · score 76
Install: claude install-skill sgtlaugh/skills
# /darkmoon - invade the last answer, judge it on evidence Blade of the Darkmoon: an invader summoned to **judge a transgression.** Here the transgression is *a possibly-wrong answer the author just gave.* darkmoon spawns an **independent, fresh-context skeptic** that tries to break the author's last response - but it can only **refute with grounded evidence** (a mechanical check, else Can't-determine), upholds what survives, and never manufactures a counter to look useful. **Why it exists.** The author cannot honestly grade its own last answer (a self-judge rationalizes - the same reason an author can't critique its own draft). darkmoon is the external check on *the author itself*: confident-wrong answers (a misread `grep`, an over-stated claim, a recommendation resting on an un-grounded premise) get caught by someone who didn't write them. **Honest ceiling.** One skeptic, grounded but fallible - "Upheld by darkmoon" **reduces** confident-wrong, it doesn't **eliminate** it (one run can miss; for high-stakes, run several passes in fresh processes and union the findings - separate processes keep each skeptic independent). And darkmoon can itself misread - so it falsifier-checks its **own** counters before emitting them. It is a sharp second opinion, not an oracle. ## Usage ``` /darkmoon # challenge the author's whole last response /darkmoon <claim> # challenge one specific claim/topic from it (cheaper, sharper) /darkmoon --context N # also include the N