The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external data by instructing the agent to analyze logread and dmesg outputs from the target device. This creates a potential surface for indirect prompt injection if the log content is maliciously crafted.
Ingestion points: logread and dmesg commands defined in SKILL.md (Workflow R) and references/uart-basics.md.
Boundary markers: The instructions do not specify any delimiters or warnings to ignore instructions within the logs, though it provides specific filters like tail -50.
Capability inventory: The skill is scoped to diagnostic command execution and calls to specific recovery scripts like armbian-install on the target device. It does not include tools for external network communication or arbitrary file system modifications on the host agent.
Sanitization: There is no evidence of sanitization or character escaping for the log data before it is processed by the agent.

openwrt-serial-recovery