Prevent common coding mistakes with battle-tested agent skills

nodumbmode is a collection of skills for AI agents, each one created only after a specific, concrete failure in a real project - no skill here is added for a theoretical best practice. The pack contains four skills: nodumb, a common-sense manifest and rule set against typical agent mistakes (understand the task, check its scope, and gather facts before writing code), meant for use before non-trivial code changes, mass edits, or a second debugging attempt, producing a short output of what is known, where the risk is, what to do next, and how to verify the result; changelog-discipline, which keeps a changelog as a decision log recording what changed, why, what it was before, and what alternatives were rejected; system-feedback, which checks whether it is clear to the user what happened after an action and what to do if something fails; and ask-nodumb, a design consultant for product and UX tasks that helps surface a hidden assumption and separate the product vision, the model, and the technology.

Use nodumb before starting non-trivial code work, a broad edit across a codebase, or a second attempt at debugging something that failed once already. Use changelog-discipline whenever a project needs a running record of decisions rather than just a list of changes. Use system-feedback when building or reviewing user-facing flows where error states and post-action clarity matter. Use ask-nodumb during product or UX design discussions where an unstated assumption might be driving the design.

npx skills@latest add hronicasync/nodumbmode

This requires Node.js; the CLI asks whether to install globally or into the current project, detects installed agents, and lets you choose which ones to install into - Claude Code, Codex, Cursor, Windsurf, Continue, and dozens more are supported the same way. For a non-interactive install (CI or a script), name everything explicitly:

npx skills@latest add hronicasync/nodumbmode -g -y -a claude-code -a codex --skill '*'

Skills are written once to ~/.agents/skills/, with symlinks placed into each agent's own directory, so an update reaches every linked agent at once without duplicating files. Updating specific skills:

npx skills@latest update nodumb ask-nodumb changelog-discipline system-feedback

Running skills update with no names updates every installed skill, including ones from other sources, which is why this pack's update command names its skills explicitly.

Works with any agent harness supported by the skills CLI, including Claude Code, Codex, Cursor, Windsurf, and Continue. SKILL.md in each skill's folder holds the core rules; the references/ folder holds examples and further detail.

Developers who want an agent's working habits (task scoping, changelog discipline, clear user feedback, questioning hidden assumptions) shaped by a record of real failures rather than generic advice.