The Agent Skills Directory

[COMMAND_EXECUTION]: The bundle includes several maintenance and validation scripts that use subprocess.run to execute git commands (e.g., git ls-files, git show). These are used for internal repository checks (secrets scanning, file synchronization) and do not pose a risk to the runtime environment.
[INDIRECT_PROMPT_INJECTION]: The skills ingest data from an external API (mcp.carrentallist.com). This attack surface is proactively mitigated by a dedicated security reference (references/untrusted-content.md) which instructs the agent to treat all fetched content as data, never as instructions, and provides specific rules for ignoring steering attempts within the data.
Ingestion points: External data is fetched via lib/carrentallist_client.py from the CarRentalList API.
Boundary markers: The skill uses references/untrusted-content.md to establish strict boundaries for external content.
Capability inventory: The bundle performs network operations (HTTPS POST) and local script execution for data fetching.
Sanitization: The agent is explicitly instructed to analyze or reply to data without following embedded instructions.
[EXTERNAL_DOWNLOADS]: The lib/carrentallist_client.py tool communicates with mcp.carrentallist.com via HTTPS to retrieve live car rental listings. This is the primary intended function of the skill and targets the vendor's own API.
[PROMPT_INJECTION]: Static analysis flags regarding concealment in references/renter-checklist.md are false positives; the file contains a 'Red flags' section that actually instructs the agent to transparently warn the user about potential rental scams (e.g., requests for Zelle or crypto payments).

exotic-car-rental