The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external documents (e.g., ticket.md, research files) from the .humanlayer/tasks/ directory. If these files contain data from untrusted sources, they could provide malicious instructions that influence the agent's actions during the design phase.
Ingestion points: Documents are read from the task artifact directory using the Read tool as specified in the instructions.
Boundary markers: The instructions do not define clear boundaries or markers to separate untrusted document content from the agent's execution context.
Capability inventory: The skill can execute Bash commands for directory listing, Read and Write files, and trigger specialized sub-tasks using Agent() calls.
Sanitization: No validation or sanitization process is described for the content of the ingested files.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute file system commands (e.g., ls -La) to correctly traverse directory structures and identify task-related artifacts.

create-design-discussion