The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect injection by processing untrusted user prompts for cost estimation and logging purposes.
Ingestion points: The prompt string is ingested by CostAwareKlingClient.text_to_video and the UsageTracker.log method in SKILL.md.
Boundary markers: While the instructions mandate a redaction policy and aggregate-only logging, the code implementation lacks explicit delimiters or boundary markers to isolate untrusted input.
Capability inventory: The skill is configured with the Write tool for local logging and utilizes the requests library for communication with the Kling AI API.
Sanitization: The UsageTracker implementation captures a 100-character preview of the prompt without filtering or sanitization, creating a minor discrepancy with the instruction to never store prompt previews.
[DATA_EXFILTRATION]: The UsageTracker class in SKILL.md implementation logs prompt fragments to a local JSONL file (kling_usage.jsonl). Although these logs remain on the local file system, they present a potential data exposure risk for sensitive information contained within user prompts, even though the skill's instructions explicitly warn against storing such data.

klingai-cost-controls