What problem does it solve? Manually probing a conversational engine for parsing and inference failures is slow and inconsistent. This Skill automates a repeatable playtest loop that walks realistic six-area conversations against a local tmct build, finds edges (unparseable requests or wrong answers), minimizes them, fixes them, and ships each iteration with a version bump. ## Core Features & Use Cases - Structured six-area playtest loop: Walks personal introductions, capability orientation, codebase exploration, teach-and-infer, code-task planning, and research digests in one continuous session per iteration. - Edge minimization and regression testing: Reduces each failing probe to a minimal passing/failing pair, diagnoses the responsible interpretation layer, and ships a named regression test with every fix. - Logged, versioned iterations: Writes a numbered playtests/PLAYTEST_LOG_*.md per run, commits the fix plus log, bumps the patch version, and pushes every iteration. - Use Case: A maintainer says "run 5 playtests" and the Skill serially probes the local tmct CLI against a scratch copy of an example repo, fixes discovered grammar or construction-bank gaps, and ships five versioned commits. ## Quick Start Tell the session to follow the playtest-edge-hunt skill and run a chosen number of playtest iterations against the local tmct build.
