The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates remote shell execution on a shared host via SSH. This is the primary function of the deployment scripts (deploy-release.sh and release-control.sh). Security is maintained through several layers:
Use of printf -v cmd '... %q' to properly escape all variables passed to the remote bash environment, preventing command injection.
Implementation of a validate_relative_path function that rejects absolute paths, parent directory traversal (..), and unsafe characters.
Strict regex validation for all input variables including RELEASE_ID, SSH_HOST, SSH_PORT, and SSH_USERNAME.
[INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface for untrusted data as it processes environment variables and GitHub event data to construct deployment commands. However, the risk is mitigated by:
Mandatory evidence chain: Ingestion occurs via environment variables in GitHub Actions; strict boundary markers are implemented using shell quoting; capabilities include file and shell operations on remote hosts; sanitization is performed via regex and path validation functions.
The instructions explicitly warn against interpolating untrusted event fields directly into shell commands and mandate passing them through validated environment variables.
[DYNAMIC_EXECUTION]: The deployment scripts dynamically assemble shell scripts to be executed on the remote host via ssh. While dynamic code generation is usually a concern, here it is used according to industry best practices for deployment automation, employing set -euo pipefail and heredocs to ensure predictable and secure remote execution.
[CREDENTIALS_SAFE]: The skill handles SSH private keys and host keys correctly by creating temporary files in restricted directories (umask 077) and ensuring they are cleaned up immediately after use via trap cleanup routines.

php-shared-hosting-deployment