The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill triggers actions and configuration overrides based on untrusted data from repository files, commit messages, and PR titles. While standard for CI/CD, these are ingestion points for external instructions.
Ingestion points: Build settings in netlify.toml (specifically command and environment keys), Git commit messages (using [skip ci] or [skip netlify]), PR/MR titles, and the USAGE.md file processed during the fullConfiguration flow.
Boundary markers: The instructions do not define specific delimiters or "ignore instructions" markers for the agent when processing commit messages or PR titles.
Capability inventory: The skill allows the agent to execute shell commands via the Netlify CLI and run arbitrary build commands defined in the netlify.toml file. It also facilitates file system modifications such as updating .gitignore.
Sanitization: No specific sanitization or validation logic is described for the content of commit messages or configuration strings processed by the agent.
[COMMAND_EXECUTION]: The documentation explicitly guides the agent to execute powerful shell commands via the Netlify CLI (netlify deploy, netlify init, netlify link, etc.) and user-defined build scripts. The skill includes safety "house rules" to restrict specific dangerous behaviors like automated rollbacks, which mitigates some of the risk associated with this capability.

netlify-deploy