The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill lists dependencies on @bunli/core and @bunli/utils in the cli/package.json file. These are not standard, widely recognized libraries, and their provenance is unverifiable, presenting a supply chain risk through potential dependency confusion or malicious updates.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted transit data and service alerts from the Rejseplanen API, creating a surface for indirect prompt injection if the API response contains instructions aimed at the agent.
Ingestion points: Untrusted data is retrieved from https://www.rejseplanen.dk/api via the apiFetch helper used in all command modules (e.g., location.ts, trip.ts, disruptions.ts).
Boundary markers: The output lacks delimiters or instructions to ignore potentially malicious embedded content when presenting API data to the agent.
Capability inventory: The agent is granted access to the Bash tool to execute the CLI, which increases the impact if the agent is manipulated by malicious data.
Sanitization: Data fields such as station names and disruption messages are passed directly to the output without sanitization or validation against prompt injection patterns.

rejseplanen