A permitting department wants every coding-agent run in Antigravity to apply the same code-review standard, covering naming conventions and error-handling patterns, without re-explaining the standard in every session. Which Antigravity customization primitive best packages this reusable review standard so the agent can invoke it whenever a matching review task comes up?
Select an answer to reveal the explanation.
Short Explanation
A skill is like a laminated checklist you hand the agent only when the job at hand actually calls for it, instead of taping instructions to every single thing it does. Naming conventions and error-handling patterns are a specific kind of know-how, so packaging them as a skill lets the agent pull that checklist out for review tasks and leave it in the drawer otherwise. That keeps the standard reusable and easy to update in one place, without re-explaining it session after session.
Full Explanation
The mechanism is scoped, invocable knowledge: a skill packages procedural expertise, such as a review checklist, that the agent recognizes as relevant to a specific kind of task and applies on demand, which is exactly the fit for a review standard that should surface during code-review work without governing everything else the agent does. A rules file operates at a broader layer — it constrains behavior across the entire repository at all times, which is the right primitive for blanket conventions like commit-message format but heavier than needed for a standard meant to activate only during review. An MCP server is built to expose external tools or data sources the agent calls out to, not to hold static internal guidance that never changes based on a live backend, so routing review criteria through one adds unnecessary infrastructure. A hook fires automatically around a specific lifecycle event, such as a commit, which enforces a gate rather than supplying the review judgment itself — useful for blocking bad commits, not for teaching the agent what "good" looks like. The caveat: if the standard later needs to be enforced rather than merely applied, pair the skill with a hook. Operational check: confirm the skill is listed as available and gets invoked in the agent's session log during a sample review run.