feat(cli)!: Default commands to global scope

[gricha]

Summary

• Make unqualified scope-aware commands operate on global state, with --project selecting repository-local state.
• Preserve --global and legacy --user aliases, while rejecting project/global conflicts before command execution.
• Centralize scope resolution, generate explicit-project hooks, and safely diagnose and repair legacy managed hooks.
• Update the changelog, release guidance, public docs, and Skillet-generated dotagents/QA skills for the v3 migration.

Migration and release

Repository workflows must add --project to every scope-aware command. Existing files are not copied, merged, or removed automatically. Legacy hooks can be updated with dotagents --project doctor --fix.

Package manifests intentionally remain at 2.2.0; Craft must publish this change with a major release. If rollback is necessary, restore the v2 npm release to the latest dist-tag.

Validation

• pnpm check (291 library tests, 830 CLI tests)
• pnpm qa:example
• packed-package verification for both workspace packages
• non-root Docker QA for global/project isolation, aliases and conflicts, non-Git project init, help output, and legacy hook repair
• Skillet hosted status/validation, dry evals, and focused changed-scenario evals for both bundled skills

Review focus

Please focus on the centralized scope boundary in cli/main.ts and cli/context.ts, plus managed post-merge hook replacement and migration behavior.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
dotagents	Ready	Preview	Aug 11, 2026 4:57pm

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit e028f45. Configure here.}

[sentry-warden]

Interactive init double-calls runInit with empty agents array

After a user completes interactive init, the outer function now falls through and calls runInit with an empty agents array, which can overwrite the agents and trust settings they just selected.

Evidence

• runInteractiveInit already calls runInit internally with the interactively-chosen agents and trust values.
• The old code guarded the outer runInit call with an else branch, so it only ran in non-interactive mode.
• The new code removed the else, so the outer runInit({ agents: [] }) executes unconditionally after runInteractiveInit returns.
• Because no trust value is passed in the outer call, the second invocation can also overwrite the trust configuration established during the interactive prompts.

_{Identified by Warden · code-review · FNX-RDC}