What problem does it solve? Running a full authorized CTF competition manually requires tracking dozens of challenges, routing each to the right specialist, validating candidate flags, and managing retry budgets and resume state. This Skill automates that entire loop through a contest API so no challenge is lost, double-submitted, or submitted with unverified flags. ## Core Features & Use Cases - State Machine Enforcement: Every challenge advances through new, prepared, solving, candidate, submitted, and solved states, with stuck and deferred branches and no skipping candidate checks. - Category-Based Routing: Normalized categories (web, pwn, crypto, reverse, forensics, misc) are dispatched to dedicated sub-agents for specialized solving. - Evidence-Gated Submission: Candidates must match the platform flag format, come from reproducible artifacts, and not appear in failed submission history before being submitted. - Use Case: During an authorized multi-challenge CTF event, run the dual-model runner to list unsolved challenges, prepare and start each one, validate candidates against evidence rules, submit within a 3-attempt budget, and resume cleanly after interruption via .ctf-state/state.json. ## Quick Start Ask the agent to load the contest profile, list unsolved challenges, and begin solving the lowest-cost challenge through the candidate-check and submit workflow.
