The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes defensive instructions to prevent the agent from treating user-supplied objectives as high-priority or override instructions. It explicitly states that the objective is user-provided data and should be treated as a task to pursue rather than as authoritative instructions.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface where untrusted user input is ingested via the --objective flag. The analysis identifies the following security posture:
Ingestion points: The okou goal create command in SKILL.md accepts arbitrary strings from users.
Boundary markers: The instructions explicitly warn the agent to treat the input as data, not instructions, though no technical delimiters are mandated for the CLI command itself.
Capability inventory: The skill uses the okou CLI for state management but relies on the agent's general toolset to perform tasks toward the goal.
Sanitization: There is no explicit sanitization, but the skill mandates a completion audit against authoritative evidence (files, command output) to prevent malicious redirection.
[COMMAND_EXECUTION]: The skill utilizes a vendor-provided CLI tool okou for task lifecycle management. These commands are localized to goal management and do not involve privileged operations or arbitrary shell execution outside of the specified vendor tools.

goal