What problem does it solve? When you already have a working codebase with an evaluation path but want to improve its results, it is unclear how to enter a structured research workflow without losing the original objective. This Skill establishes the correct entry point into the dsh_autoresearch pipeline, preserving whether the goal ends at a promoted method or also requires a paper. ## Core Features & Use Cases - Entry Decision Management: Confirms the workspace contains a runnable method and evaluation path, then persists entry_mode, deliverable (method or paper), and optional target_venue based strictly on the direct human request. - Baseline-First Initialization: Inspects implementation, configs, and prior results, loads baseline-assessment, and initializes canonical research state via transition_guard.py only after a baseline artifact exists. - Pipeline Handoff: Delegates the scientific work to depth-research-loop and routes to paper-writing when the guard's next_action requires it, without widening the original deliverable. - Use Case: You have a machine learning repository with training code and a benchmark script but weak results. Use this Skill to baseline the current method, then autonomously iterate on experiments until a verified improvement is promoted. ## Quick Start Use the improve-existing-project skill to autonomously improve this existing project through confirmed evidence.
