lcs-debug-ext
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted external data, including logs, stack traces, HAR files, and error messages, which are not explicitly delimited or sanitized.
- Ingestion points: SKILL.md specifies reading user descriptions, logs, traces, HAR files, and test output.
- Boundary markers: The instructions do not define specific delimiters for untrusted log data to prevent the agent from interpreting embedded instructions as its own.
- Capability inventory: The skill uses file reading, shell command execution (for reproduction tests), and local file writing for reports.
- Sanitization: There are no instructions for sanitizing or escaping content from external logs or artifacts before processing.
- [COMMAND_EXECUTION]: The workflow requires the agent to execute reproduction commands, such as "small targeted test commands" or "CLI invocations with fixture input," which could be exploited if reproduction steps are influenced by malicious input.
- [DATA_EXFILTRATION]: The agent is instructed to inspect sensitive artifacts, including environment details and project references, to diagnose failures. This access is scoped to debugging but involves reading files that may contain credentials or sensitive configurations.
Audit Metadata