pp-tessie
Warn
Audited by Gen Agent Trust Hub on Aug 27, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill implements a 'learning loop' that generates and executes 'playbooks' stored in a local SQLite database. These playbooks contain command sequences with variable placeholders that are resolved and executed at runtime, allowing the execution of logic derived from stored data.\n- [EXTERNAL_DOWNLOADS]: The skill requires the 'tessie-pp-cli' tool and provides instructions for its installation via 'npx -y @mvanhorn/printing-press-library' or 'go install' from the author's GitHub repository. These represent external dependencies necessary for the skill's operation.\n- [REMOTE_CODE_EXECUTION]: The installation methods provided (npx and go install) involve fetching and executing code from remote repositories, which is a form of remote code execution for setup purposes.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes data from the Tessie API and a local 'learnings' store, creating an attack surface where malicious data could influence agent actions via playbook steps or notes.\n
- Ingestion points: Tessie API responses, user-supplied text in 'teach' and 'recall' commands, and persistent data in the local 'data.db'.\n
- Boundary markers: No explicit delimiters or isolation mechanisms are defined for data interpolated into command templates or playbooks.\n
- Capability inventory: The agent can execute shell commands ('tessie-pp-cli'), write to local files, and transmit data to external webhooks via the '--deliver' flag.\n
- Sanitization: Documentation advises stripping PII from queries but does not specify validation or sanitization rules for the structural command data retrieved from the database.\n- [COMMAND_EXECUTION]: The skill serves as a broad interface for the 'tessie-pp-cli' binary, enabling the agent to execute diverse vehicle commands and manage API authentication.\n- [DATA_EXFILTRATION]: The skill includes a '--deliver' flag that supports a 'webhook:' sink, enabling the automated transmission of command output and vehicle data to arbitrary network endpoints.
Audit Metadata