The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret JSON and JSONL data records from external OAK hardware pipelines via the oakctl inspect dump command.
Ingestion points: External pipeline messages and data topics are read and parsed by the agent as described in SKILL.md (Step 5).
Boundary markers: The instructions lack specific guidance on using delimiters or boundary markers to isolate potentially malicious data injected into the pipeline streams from the agent's instruction context.
Capability inventory: The skill utilizes shell command execution via oakctl and file system write operations to the evidence/ directory.
Sanitization: No explicit validation or sanitization procedures are defined for the ingested JSON/JSONL data.
[COMMAND_EXECUTION]: The skill relies on the execution of the oakctl CLI tool to perform live inspection, capture frames, and dump pipeline data, which involves direct shell interaction.
[DATA_EXFILTRATION]: The skill facilitates network communication to remote hosts using WebSockets (ws://) via the oakctl --url flag. Additionally, references/instrumentation.md provides instructions for binding services to 0.0.0.0, which creates unauthenticated network endpoints that could be accessed by unauthorized peers on the same network.

luxonis-inspect