- What is sandermuller/boost-core and what problem does it solve?
- boost-core is a dev tool that lets you author AI agent skills, guidelines, and commands once in a .ai/ directory, then compiles them into the native file layout of ten different AI coding agents. It sits in the developer-experience layer of a project and tracks what it owns in a manifest, so generated files stay in sync without clobbering anything you wrote by hand.
- Do I need Laravel to use boost-core?
- No. boost-core is framework-free with no runtime dependency, so it works in Laravel apps, Symfony services, legacy PHP codebases, packages, and even non-PHP repositories. You add it as a Composer dev dependency and it publishes agent configuration for that project.
- How do I install and run boost-core?
- Run composer require --dev sandermuller/boost-core, add the config entry where you choose agents with withAgents(), then run boost sync. Sync is idempotent enough to run repeatedly after config edits, and a --check mode lets you predict problems such as skill-name clashes before anything is written to disk.
- Which AI agents does boost-core publish to?
- Ten: Claude Code, Cursor, GitHub Copilot, Codex, Gemini, Junie, Kiro, OpenCode, Amp, and Antigravity. You control the list in withAgents(), and adding an agent means its files simply appear on the next sync with no per-agent copy to maintain.
- Will boost-core overwrite my existing CLAUDE.md or other hand-written agent files?
- It tracks ownership in a manifest, so adopting boost-core in a repo that already has a CLAUDE.md will not wipe it. User-scope sync additionally refuses to write over a file boost does not own and refuses to write through a symlink at a target path, which removes the silent-overwrite risk.
- How does tag scoping work in boost-core?
- Tag filtering lets you keep a skill out of projects where it does not apply, such as a jira-triage skill in a repo with no Jira work. It also keeps the skill out of the agent's skill-selection index, where every unwanted description costs the model attention.
- Can I distribute skills as a Composer package?
- Yes. Any package that ships resources/boost/skills/ becomes a skill source once you allowlist its vendor, so a team can author its skill set once and composer require it everywhere. Skills can also declare dependencies on other skills via metadata.boost-requires, so a hand-off like 'then run code-review' ships as a unit instead of breaking silently.
- What is the difference between project scope and user scope in boost-core?
- Project scope writes into the repository so the whole team shares one configuration. User scope, enabled with --scope=user, writes per-machine output curated by a ~/.boost/user-scope.php selector, and a -user suffix prevents a user skill from shadowing a project skill of the same name. Because user scope is not versioned with the repo, treat it as personal preference rather than team policy.
- Can I run boost-core in CI and how do I avoid breaking changes?
- Sync is idempotent and the --check mode verifies state without writing, so it fits a CI pipeline. Generated paths can change when an agent's expected layout turns out to be broken, so pin an exact version and read UPGRADING.md before bumping, especially for user-scope output.
- Is boost-core actively maintained and does it follow semantic versioning?
- The project ships a structured changelog and a dedicated UPGRADING.md, and consecutive releases indicate an active, disciplined cadence, but its adoption base is still small. Exact version pinning is wise, since a minor release has already changed generated user-scope paths when the old nested layout silently made skills invisible to Claude Code.