The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents an indirect prompt injection surface as it ingests data from a booking record (including booking IDs and node IPs) produced by the ito-compute skill. This data is interpolated into shell commands for the ecc tool without documented sanitization or boundary markers in the instructions. While this is an intended operational workflow, it depends on the integrity of the data provided by the preceding skill.
[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands using the ecc CLI utility. These commands are used to delegate inference tasks to the canonical backend. The skill correctly identifies that authentication is handled via device authorization, avoiding the exposure of secrets in command-line arguments or log files.
[SAFE]: The skill identifies itself as a scaffold, stating that current CLI verbs for inference are not yet fully implemented. It explicitly instructs the agent to stop and report the missing capability rather than attempting to substitute local runners or unauthorized endpoints.

ito-inference