What problem does it solve? Complex tasks often fail because agents skip planning, guess at facts mid-execution, or ship unverified changes. This Skill enforces a disciplined four-stage loop—plan, execute, verify, audit—so multi-step work is evidence-driven and adversarially checked before delivery. ## Core Features & Use Cases - Parallel Evidence Fan-Out: Spawns concurrent research and codebase-exploration workers that return distilled, cited findings before any plan is committed. - Adversarial Verification: Launches 1-3 attacker workers that attempt to refute the change from distinct lenses, catching defects the main thread missed. - Gated Execution: Applies intent, recall, and authorization gates so edits stay minimal, facts are never recalled from memory, and outward-facing actions require explicit user approval. - Use Case: When asked to refactor an authentication flow across a codebase, the loop gathers evidence on all call sites in parallel, commits to one plan, executes surgically, then has attacker agents try to break the change before reporting the verified outcome. ## Quick Start Ask the agent to run the fable loop on your next non-trivial task, for example by saying run the fable loop to migrate the settings page to the new API.
