The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/summarize_tests.py utilizes the xml.etree.ElementTree module to parse GTest and JUnit XML reports. This library is documented to be insecure against maliciously constructed data. Specifically, it does not prevent XML External Entity (XXE) expansion, which could be exploited by an attacker providing a crafted test report to read arbitrary local files or perform server-side requests.
Evidence: ET.parse(path) in scripts/summarize_tests.py.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize test reports from external or potentially untrusted runs (e.g., from pull request artifacts). Malicious content embedded in test names, class names, or failure messages within the XML could be used to influence the agent's analysis or misrepresent the state of the codebase.
Ingestion points: scripts/summarize_tests.py (reads test report files specified in CLI arguments).
Boundary markers: None; the script extracts text content directly from XML nodes.
Capability inventory: The skill can trigger GitHub workflows, read system process information, and execute local scripts.
Sanitization: No formal sanitization or escaping of extracted XML content is performed before generating the JSON summary.
[COMMAND_EXECUTION]: The skill instructions direct the agent to perform various command-line operations, including running local Python scripts for process introspection and triggering GitHub Actions workflows.
Evidence: Documentation in SKILL.md and references/validation.md specifies usage of gh workflow run, python runtime_identity.py, and python summarize_tests.py.

ov-gguf