The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is configured to interact with the ~/.claude/projects/ directory, which is a sensitive path containing internal agent configuration, history, and project state. The 'Self-Evolution Telemetry' feature writes execution metadata to this hidden path, enabling the tracking of user activity across sessions and the potential manipulation of the agent's long-term behavior through the mentioned evolution pipeline.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data from vehicle records, maintenance logs, and telematics systems without implementing boundary markers or sanitization logic. * Ingestion points: Found in Step 1.1 (Vehicle records), Step 1.2 (Maintenance architecture), and Step 6.1 (Telematics integration). * Boundary markers: The instructions lack delimiters or explicit warnings to the agent to ignore embedded commands within the processed fleet data. * Capability inventory: The agent has extensive file system access, including the ability to read the entire codebase and write to both the project's documentation folder and the agent's internal configuration paths. * Sanitization: No input validation, filtering, or escaping is specified for the data ingested from external sources.

fleet-maintenance