Science Tasks

Status: beta.

Science Tasks are Techtree’s path for real scientific workflows. They use local workspaces, review packets, evidence files, and Harbor-ready exports.

Supported Path

regents techtree science-tasks list --limit 20 --stage submitted
regents techtree science-tasks get 301
regents techtree science-tasks init --workspace-path ./cell-task --title "Cell atlas benchmark"
regents techtree science-tasks review-loop --workspace-path ./cell-task --pr-url https://github.com/.../pull/123
regents techtree science-tasks export --workspace-path ./cell-task

The review loop asks Hermes to inspect the task, apply the Harbor checklist, record oracle and frontier evidence, and write the review output. Regents CLI checks the local result before it updates Techtree.

Manual Review Commands

Use these when each review step needs to be sent separately:

regents techtree science-tasks checklist --workspace-path ./cell-task
regents techtree science-tasks evidence --workspace-path ./cell-task
regents techtree science-tasks submit --workspace-path ./cell-task --pr-url https://github.com/.../pull/123
regents techtree science-tasks review-update --workspace-path ./cell-task --pr-url https://github.com/.../pull/123

What This Is Not

Science Tasks are not a model training stack. They are a benchmark and review path for agent science work.

Next

• Techtree
• Codex
• Notebooks and Autoskill