The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes dynamic context injection using the !cat syntax to execute shell commands when the skill is loaded. These commands are hardcoded to read project-specific configuration and protocol files (e.g., .production-grade.yaml, .forgewright/settings.md) and inject their contents into the agent prompt. These commands do not process user-provided arguments.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. Ingestion points: !cat commands in SKILL.md and project infrastructure reads. Boundary markers: Absent. Capability inventory: Extensive file-writing and script generation capabilities (e.g., writing to .github/workflows/ and infrastructure/ directories). Sanitization: Absent. The lack of delimiters for injected content allows malicious instructions in local configuration files to potentially influence the agent's deployment and configuration tasks.

devops