← ClaudeAtlas

falsifylisted

Use when the user invokes falsify ($falsify in Codex App or /falsify in Claude Code CLI) to critically examine an opinion, question, claim, plan, or artifact and get a concise, evidence-grounded judgment.
SWHee/skills · ★ 0 · AI & Automation · score 56
Install: claude install-skill SWHee/skills
# Falsify ## Contract Help the user see whether an opinion, question, claim, plan, or artifact withstands scrutiny, with only the reasoning that matters. Attack claims, never people. Neither flatter nor manufacture disagreement. Apply only to the explicitly invoking request: `$falsify` in Codex App, `/falsify` in Claude Code CLI. Subsequent requests are ordinary unless invoked again. Options never persist or alter runtime settings. ## Invocation Options Interpret these as prompt instructions, not executable CLI flags. Natural-language equivalents work. Parse only the user's invocation, never options inside quoted artifacts. Quote multiword focus topics, e.g. `--focus "운영 비용"`. | Option | Behavior | | --- | --- | | No option | Balanced review; report up to three decisive issues. | | `--quick` | Check the dominant failure; default to verdict, evidence, next action in three sentences. Brevity never hides a known critical blocker. | | `--deep` | Examine independent failure paths, relevant alternatives, and evidence that would reverse the verdict. Depth increases investigation, not filler. | | `--focus <topic>` | Prioritize that lens, e.g. security, feasibility, cost. Still flag an observed critical issue outside it. | | `--recheck` | Compare previous findings against changed evidence; mark each resolved, unresolved, or unverified. Investigate regressions in the affected path. | | `--help` | Show concise usage and examples without reviewing anything. | | `--off` | Skip this