Claude Skills by Emlembow

Run a file-audited completion loop that requires two consecutive fresh-context subagents to independently try to disprove the same versioned, digest-checked task result. Use when the user invokes $adversarial-review, asks for two independent adversarial approvals, or requires a result to survive repeated attempts to find correctness, completeness, requirement, or verification gaps before completion. Do not trigger for an ordinary one-pass review, and never self-certify when fresh subagents ar...

Coordinate complex tasks through verified completion using subagents, an adaptive dependency plan, durable task records, and steady progress readouts. Use when explicitly invoked as $divisible-work or when a complex user task warrants sustained planning, delegation, monitoring, and integration, including research, implementation, and existing task lists. Decide what can run in parallel and revise the path as results arrive. Do not trigger for simple requests, explanations of Divisible Load Th...

Run autonomous, metric-driven experiments on a version-controlled implementation against a fixed trusted evaluation harness. Use when the user asks to improve eval pass rate, benchmark score, prompt or policy quality, performance, cost, or another measurable outcome through repeated hypothesis, change, evaluate, and keep-or-discard cycles. Protect generalization with holdout gates, immutable evidence provenance, and replay-safe recovery. Reject hardcoded cases, benchmark leakage, scenario-spe...

State what this skill does, the prompts or contexts that should trigger it, and any nearby tasks that should not trigger it.