piebird-workflow-rules
Rules and a skill on how to work, for any project: weigh problems in proportion, think before building, fix a correction as a class of mistakes, use the project's own tools, keep the changelog, and ask only for real decisions.
omp, Claude Code and Codex fix the same reported bug under the same rules
Install
omp plugin marketplace add georgelubaretsi/piebird
omp plugin install piebird-workflow-rules@piebird
Claude Code: claude plugin marketplace add georgelubaretsi/piebird, then
claude plugin install piebird-workflow-rules@piebird. Codex: codex plugin marketplace add georgelubaretsi/piebird, then codex plugin add piebird-workflow-rules@piebird. ChatGPT and
Claude on a Mac or a Windows PC: the piebird installer (Install).
What it adds
| Rule | How omp loads it |
|---|---|
proportion, think-before-building, recurring-corrections | in full, every agent |
proportion-review | in full, the reviewer agent |
ask-for-decisions | in full, the main agent |
project-tooling, changelog | listed; the agent reads one when it applies |
Claude Code has no rules of its own, so the plugin's mod puts them in a block of the conversation's context: the always-applied ones for the main agent in full, the others by name and file; the reviewer's rule stays out. Codex gets the skill.
The skill failure-classes lists ten general failure classes, each with its members and how to
test them.
What the model sees: every rule and the skill, word for word, and the block Claude Code gets.
Codex code review
Codex code review doesn't load plugins. To give it the same bar for findings as the reviewer
agent, paste the text of rules/proportion-review.md (below its frontmatter) into ChatGPT
Settings > Code Review, and paste it again when this plugin's changelog says that rule changed.
OpenAI documents the setting for reviews started in the ChatGPT app; check that your GitHub
reviews follow it too.