tech-stack-evaluator
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The script
scripts/evaluate.pyreads file content based on the path provided in its command-line arguments usingPath(path).read_text(). This capability could be abused to expose sensitive local files if the agent is manipulated into targeting an unintended file path. - [COMMAND_EXECUTION]: The skill requires the agent to execute Python scripts via the shell. One module,
scripts/report_generator.py, contains anexport_to_filemethod that writes content to arbitrary file paths, posing a risk of overwriting critical system files or configurations. - [DYNAMIC_EXECUTION]: The main entry script
scripts/evaluate.pyusesimportlib.utilandexec_moduleto load its component libraries at runtime. While the names of the modules are currently hardcoded, this dynamic loading pattern is a sensitive execution vector. - [INDIRECT_PROMPT_INJECTION]: The skill's core workflow requires the agent to research technologies online and incorporate 'evidence' from external websites and repositories into its scores. This creates an indirect prompt injection surface where malicious instructions in third-party documentation could influence the agent's actions. Ingestion points: Command-line input files (JSON/Text) and external web content from research. Boundary markers: Absent; no specific delimiters or instructions to ignore embedded commands are employed. Capability inventory: The agent executes local Python scripts that have arbitrary file read/write access. Sanitization: Absent; the skill processes text content without specific security filtering.
Audit Metadata