The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to run npx unbox-ai, which fetches the unbox-ai package from the npm registry at runtime if it is not already available locally.
[REMOTE_CODE_EXECUTION]: Executing an external package via npx constitutes remote code execution as it involves fetching and running binary content from an external registry.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process AI agent trace files that may contain untrusted data including prompts and tool outputs, creating a surface for indirect instructions.
Ingestion points: The skill processes *.trace.json and .devtools/generations.json files as identified in SKILL.md.
Boundary markers: The skill description in SKILL.md mentions that the tool provides "bounded output," but it does not specify the use of delimiters or instructions for the agent to ignore commands embedded within the trace data.
Capability inventory: As specified in SKILL.md, the skill has the capability to execute CLI commands and read local files.
Sanitization: The skill notes that the tool "normalizes" traces, but there is no mention of sanitizing or escaping potentially malicious instructions embedded in the message content of the traces.

unbox-ai