error-diagnostics-error-trace
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The ErrorTrackingAnalyzer class in resources/implementation-playbook.md ingests untrusted data by reading source code files from a user-provided project path.
- Ingestion points: ErrorTrackingAnalyzer._analyze_error_handling reads the content of all files matching specific extensions (.js, .ts, .py, .java, .go) in the project directory using Path.rglob and read_text.
- Boundary markers: The skill does not implement delimiters or explicit instructions to ignore embedded prompts within the analyzed code files.
- Capability inventory: The script uses ast.parse for static analysis and regex for pattern matching. While it does not execute the code or perform network operations itself, the results are processed by the agent to generate reports.
- Sanitization: The ErrorTracker class provides a sanitizeEvent method that redacts sensitive keys (e.g., password, token, secret, api_key), but the primary code analysis script does not sanitize the input text.
- [COMMAND_EXECUTION]: The RecoveryManager and CircuitBreaker templates in resources/implementation-playbook.md execute arbitrary functions passed via the operation argument. While this is a standard design pattern for retry logic and circuit breakers in software engineering, it establishes a mechanism for functional callback execution.
- [SAFE]: The skill follows security best practices by including logic to redact sensitive information from logs before they are transmitted to external services, as seen in the SentryErrorTracker (cookies filtering) and ErrorTracker (sensitive key redaction) implementations.
Audit Metadata