The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions (SKILL.md) direct agents to use shell commands such as echo, cat, and tail to record and verify the status of parallel tasks within the local .claude/cache/ directory.
[PROMPT_INJECTION]: The orchestration template is susceptible to indirect prompt injection because user-defined variables are used directly in shell commands.
Ingestion points: Subagents receive <identifier> and <batch> names which are interpolated into echo commands in SKILL.md.
Boundary markers: There are no delimiters or instructions to ignore embedded characters in the interpolated variables.
Capability inventory: Subagents have the capability to execute shell commands via the Task interface.
Sanitization: The skill does not provide methods to escape or validate these variables before execution.

parallel-agents