What problem does it solve? Complex tasks with multiple dependent steps, unknowns, or debugging often fail because the first approach is wrong and discipline degrades over long sessions. This Skill enforces a structured working method so hard tasks are scoped, evidenced, attacked, verified, and reported rigorously. ## Core Features & Use Cases - Five-Gate Task Loop: Scope before work, gather evidence before reasoning, reason adversarially, verify before declaring done, and report with calibrated confidence. - On-Disk State Persistence: Maintains .work/GATES.md and a marker file so method discipline survives long contexts and session restarts. - Adversarial Review Barrier: Requires an independent cold-context reviewer to attack the answer before verification, with explicit disposition of every finding. - Use Case: When debugging a failing deployment where two fixes have already failed, activate fable mode to re-scope the problem, probe the load-bearing unknowns, get an independent review of the diagnosis, and verify the fix against the original check before reporting. ## Quick Start Tell the AI to use fable mode to work through this debugging task step by step and verify the fix before reporting done.
