The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The scripts/acceptance.sh script contains a command to execute a sudo reboot on the target host via SSH. While this is documented as an intentional recovery test, it constitutes a high-privilege system operation initiated by the skill. Evidence: ssh "$ssh_alias" 'sudo -n systemd-run --on-active=2s /usr/sbin/reboot'.
[REMOTE_CODE_EXECUTION]: The green launcher script dynamically fetches and executes Clojure code from external GitHub repositories during its execution. Evidence: The script uses babashka.deps/add-deps to pull dependencies from github.com/getcolors/dbos.git, github.com/getcolors/once.git, and github.com/getcolors/green.git using specific git commit SHAs.
[DYNAMIC_EXECUTION]: The launcher employs dynamic resolution to invoke functions from libraries that are not present at the start of execution but are loaded during the bootstrap phase. Evidence: The use of requiring-resolve in green to call entry points like green.cli/run-cli.
[COMMAND_EXECUTION]: The skill modifies the user's local ~/.ssh/config file and manages SSH keypairs in the ~/.ssh/ directory to enable connectivity to the deployed cloud instances. Evidence: Detailed logic for block insertion and key generation described in references/configuration.md and implemented in the launcher.
[INDIRECT_PROMPT_INJECTION]: The scripts/acceptance.sh script parses and acts upon JSON responses from a remote endpoint. This introduces a vulnerability surface where a compromised or malicious server could return data designed to influence the script's behavior. Ingestion points: Network requests via curl to $base/health and $base/workflows/$id in scripts/acceptance.sh. Boundary markers: None identified. Capability inventory: Execution of SSH commands, sudo access, network operations, and local file modifications. Sanitization: The script uses jq to select specific JSON fields, providing structural validation but not sanitizing the content for downstream consumption.

package-dbos-green