What problem does it solve?
Long-running research and engineering tasks lose context across sessions: experiment commands, results, failures, and decisions get scattered or forgotten. This Skill provides a durable, append-only logbook format plus rules for keeping a coordinating GitHub issue in sync, so work remains reproducible and auditable.
Core Features & Use Cases
- Structured Logbook Template: Creates Markdown logbooks under
.agents/logbooks/<topic>.md with YAML frontmatter, scope, baseline, and timestamped entry logs recording hypothesis, commit hash, command, config, result, and next action.
- Living Indices: Maintains editable summary sections (TL;DR, hypothesis queue, decision log, negative results index) with traceable links to entries, commits, and W&B runs.
- Coordinating Issue Workflow: Defines promotion rules for posting milestone updates as GitHub issue comments, maintaining the issue body as the current summary, and closing with a final agreed conclusion.
- Use Case: While running a multi-day model training experiment, log each run's command, hash, and metrics in the logbook, post milestone updates to the tracking issue, and close it with a final summary of what worked and what did not.
Quick Start
Create a task logbook for my current experiment and record the baseline results with the exact command and commit hash.
