What problem does it solve? Research findings often live in scattered notes or chat logs, disconnected from the experiments that produced them, and negative results get lost entirely. This Skill attaches a structured, searchable write-up — Markdown with LaTeX math and embedded figures — directly to a Chronicle experiment or variation, with a mandatory "What didn't work" section so failed approaches stay on the record. ## Core Features & Use Cases - Structured report authoring: Drafts a write-up with required sections (Summary, What worked, What didn't work, Open questions, Figures) rendered inline in the Methodic UI with MathJax. - Figure asset management: Uploads vector plots as grouped SVG/PNG/PDF assets and interactive HTML diagrams, embedding them by asset id so each surface (browser, Overleaf, Slack) gets the right format. - Citations and findings: Registers DOI/arXiv publications as citation links on the experiment and records a structured per-variation finding (working / partial / not_working) that feeds the experiment's running summary. - Use Case: After a training run finishes, ask the agent to pull the real metrics from W&B, render a loss curve, and attach a takeaways report to the experiment — complete with the ablations that regressed and citations to the baseline papers. ## Quick Start Ask the agent to write up the findings from the latest variation of the current experiment and attach the report with its loss-curve figure to the Chronicle experiment.
