The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches geolocation and routing information from the Amap (Gaode) official REST API (restapi.amap.com), which is a well-known map service. It also provides instructions for optionally integrating other common travel information sources via trusted installation methods.
[COMMAND_EXECUTION]: Utility scripts such as scripts/validate.py and scripts/render_html.py are included to process travel data and generate outputs. These scripts run locally using the Python interpreter and are designed for data integrity checks and formatting.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes content retrieved from external social media (Xiaohongshu) and guide platforms (Meituan). It effectively mitigates this risk by validating all ingested data against a strict JSON schema and applying robust escaping to content before rendering it for the user. Evidence chain:
Ingestion points: External data enters the agent context through tool outputs and is stored in trip.json (documented in SKILL.md and references/data-sources.md).
Boundary markers: The skill enforces a formal data structure using references/trip-schema.json to isolate external content.
Capability inventory: Subprocess calls are limited to local validation scripts, and network operations are restricted to established service providers.
Sanitization: Third-party content is passed through esc() and mdLite() functions in assets/template.html and render_markdown.py to prevent script injection or layout disruption.

travel-planner