The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill inherently processes large amounts of untrusted data from the target project, including XML sitemaps, HTML snapshots, and PHP source code. It also interacts with the TYPO3 backend UI. This surface is mitigated by a strict 'trust hierarchy' (defined in rules/00-scope-and-prohibitions.md) that prevents data from being interpreted as instructions, an egress proxy that restricts browser communication to the site under test, and a logic design that keeps untrusted text entirely out of verdict-producing code paths.
[COMMAND_EXECUTION]: The skill makes extensive use of system commands such as ddev, git, mysql, php, and rsync to manage the upgrade process. All commands are redacted for secrets before being written to an append-only journal, and many are subject to pre-execution checks and database snapshots to ensure reversibility.
[EXTERNAL_DOWNLOADS]: The skill includes functionality to pull database and file datasets from live production servers via SSH and to fetch sitemaps and images from the site being tested. These operations are strictly guarded by a URL allow-list, are read-only on the remote server, and require explicit host configuration and user authorization.

typo3-upgrade-run