The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the vehicle-safety-pp-cli and vehicle-safety-pp-mcp tools. These are fetched from the author's repositories using npx -y @mvanhorn/printing-press-library and go install github.com/mvanhorn/printing-press-library/....
[DATA_EXFILTRATION]: The CLI tool includes a --deliver webhook:<url> feature that allows the output of any command to be transmitted to an arbitrary external URL via HTTP POST. If an agent is directed to use this flag, it could result in the exfiltration of vehicle data, decoded VIN information, or local file contents.
[DYNAMIC_EXECUTION]: The skill uses an 'Automatic learning' mechanism where the vehicle-safety-pp-cli journals sessions to a local database and synthesizes playbook_candidate objects. The agent is instructed to follow these dynamically generated playbooks in future sessions, which involves replaying command sequences generated at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data sources that could contain malicious instructions.
Ingestion points: External files provided via --garage (CSV) and --communications-file (TSV), as well as user-supplied VIN strings.
Boundary markers: There are no instructions for the agent to use delimiters or ignore instructions potentially embedded within these data sources.
Capability inventory: The skill has access to the Read Bash tool, file writing via --deliver file:, and network transmission via --deliver webhook:.
Sanitization: The skill describes a process for stripping personal identifiers (PII) from 'taught' queries but does not mention sanitization for data ingested from files that could influence the generation of playbooks or the execution of commands.
[CREDENTIALS_UNSAFE]: The documentation explicitly points to credentials.toml within the tool's data directory as the storage location for secrets, which is accessed by the tool during its operation.

pp-vehicle-safety