Skill UI

Autonomous experiment loop that optimizes any file by a measurable metric. The agent edits a target file, runs a fixed evaluation, keeps improvements (git commit), discards failures (git reset), and loops indefinitely. Ideal for optimizing code speed, reducing bundle size, improving test pass rates, enhancing prompts, or boosting content quality through iterative measured attempts.

Allows users to resume an interrupted or paused autoresearch experiment. It automatically loads the full context, including configuration, strategy files, complete results history, and recent code changes. The system then summarizes the current best metrics and patterns, prompting the user to continue with a single iteration, set up an automated loop, or manually review the results.

This skill initializes a new autoresearch experiment. It guides the user through defining all necessary parameters, including the domain, target files, evaluation commands, metrics, and optimization direction. It is the essential starting point for automated optimization and performance benchmarking cycles.

This skill orchestrates end-to-end autonomous AI research projects using a sophisticated two-loop architecture. It manages the entire research lifecycle, from initial literature review and hypothesis formation to running constrained experiments and synthesizing findings into publishable papers. It acts as a project manager, autonomously driving the process while structuring the workspace and presenting progress to human users.

Runs a SHA-pinned integrity-forensics sweep on a paper (evidence ledger, auditor proposals, rules-only report), converting verdicts into typed policy gates (BLOCK/WARN/NO_NEW_BLOCKER) and an obligations ledger. Used for pre-submission integrity checks, peer-review forensics and resubmission pipelines.

An autonomous system designed to continuously improve code quality by scanning codebase metrics. It functions as an agent, proposing mechanical code changes, executing them, and committing only if defined metrics improve. It supports multiple modes (Scan, Scaffold, Run) for systematic, data-driven codebase refactoring.

Resume a paused or interrupted autoresearch experiment by loading full state, reporting progress, and letting the user continue iterating with /ar:run or /ar:loop. Ideal for context-limited sessions when users need to pick up prior work seamlessly.

Displays a dashboard of all autoresearch experiments, showing results, active optimization loops, and progress. Supports per-experiment detail, domain filtering, and format export (Markdown/CSV) for sharing. Useful for monitoring ongoing experiments.