How-to guides

Problem-oriented recipes. Each guide states its starting assumptions, gives a copy-paste example, and tells you the expected result (including the exit code where it matters). New to GSKILL? Start with the tutorial first.

Installing skills

• Install a local skill — add a skill from a folder on disk.
• Add a skill from Git — add a skill from a Git repository with a version constraint.
• Target specific agents — install into Claude Code, Codex, Cursor, Antigravity CLI, OpenCode, OpenClaw, and/or Hermes Agent, and choose project vs global scope.
• Copy vs symlink — choose how installed content lands on disk.
• Customize a skill without forking it — patch, replace, or layer content onto an upstream skill and keep taking updates.

Reproducing & verifying

• Reproduce exactly with --frozen-lockfile.
• Work offline — restore from a warm cache without network.
• Verify integrity — re-hash installed content against the lock.
• Gate CI on drift — fail a build when skills drift or updates are available.

Lifecycle

• Update and re-lock — move within declared constraints; never edits skills.toml.
• Upgrade a skill — change the declared version in skills.toml, then resolve and install.
• Remove a skill and reclaim space.
• Sync and repair — reconcile disk with the lock; fix broken installs.
• Inspect with list, info, and check.

Operations

• Cut and verify a release — publish a version and verify its checksums, signature, and provenance.
• Manage the cache.
• Reuse skills across projects.
• Use different skill versions per project.
• Configure GSKILL.
• Set up shell completion.
• Run doctor.
• Use the TUI dashboard.
• Run GSKILL in a container: pull the image, mount a project, keep the cache.

Scripting

• Script GSKILL with --json.

---

Not yet supported

The following commands appear in the long-term design but are not implemented yet, so they have no guide: export and import. They are documented here only so you know they are planned.