Skip to content
Kitbay
Skill

doubt-driven-development

MITUpdated on Nov 1, 2026115,373GitHub stars

A skill from Addy Osmani's agent skills collection that has every significant decision checked by a reviewer with fresh context while the work is still in progress.

What the skill teaches

A skill from Addy Osmani's agent skills collection that has every significant decision checked by a reviewer with fresh context while the work is still in progress. The agent states the decision in two or three lines, cuts out the smallest piece to review along with its constraints and any contrary evidence, and sends it to a subagent asked only to find problems, without the author's reasoning. Findings are sorted into misread constraints, real fixes, accepted trade-offs and noise, and the loop stops after three rounds or when findings turn trivial. In interactive sessions it also offers a second opinion from another model's command-line tool, run read-only after the user confirms.

Capabilities

  • Send a decision to a fresh-context reviewer subagent
  • Sort review findings into fixes, trade-offs and noise
  • Offer a second opinion from another model's CLI
  • Stop the review loop after set limits