The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): Indirect Prompt Injection vulnerability through untrusted project data.
Ingestion points: The skill explicitly directs the agent to read and parse Taskfile.yml and Taskfile.yaml from the current workspace.
Capability inventory: The skill facilitates the execution of arbitrary shell commands through the task binary based on the contents of these external files.
Boundary markers: No instructions are provided to the agent to distinguish between safe automation and malicious command injection within the task definitions.
Sanitization: There is no requirement or guidance for the agent to sanitize, validate, or even inspect the underlying shell commands before executing them.
[COMMAND_EXECUTION] (HIGH): Execution of arbitrary commands from an external source.
The skill encourages the agent to 'Always prefer task commands over direct shell/language commands,' which effectively masks the underlying commands being executed, making it easier for malicious payloads in a Taskfile to go unnoticed during the 'discovery' phase.

taskfile-automation