yallalisted

# /yalla Adaptive autonomous development pipeline. It classifies the task, chooses the right engineering path, creates or resumes a GitHub issue, plans from domain/code evidence, builds in vertical tracer-bullet slices, tests through public interfaces, reviews with binary gates, captures learnings, and ships a PR. Default shipping policy: create a PR only. Do not merge unless the user explicitly asked for auto-merge in this run. `$REPO` is the target repository, resolved from `repo:` in `.claude/YALLA.md`, or auto-detected with `gh repo view --json nameWithOwner -q .nameWithOwner`. `$BASE_BRANCH` is the branch new work is cut from and PRs target, resolved from `base_branch:` in `.claude/YALLA.md` (default `main`). Build/test commands come from the `commands:` block in `.claude/YALLA.md`; `npm test` is used here only as a generic default example. > **Locating bundled files.** Reference files shown as `${CLAUDE_PLUGIN_ROOT}/knowledge/yalla/*.md` and `${CLAUDE_PLUGIN_ROOT}/agents/*.md` are the Yalla engine, bundled with this plugin — read them from there. (If Yalla was vendored into the repo with `install.sh`, those same files live under `.claude/knowledge/yalla/` and `.claude/agents/`.) `.claude/YALLA.md` is always your project's own config and lives in your repo, not the plugin. ## Operating Principle Keep the universal pipeline small. Do not make every PR carry every historical scar. For every changed workflow, define its success invariant: the workflow is not successf