The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands via the Bash tool to list and sort files in the .github/ and test-results/ directories.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface through its processing of external test data.
Ingestion points: CI log files (from .github/, Saved/Logs/, or user-provided paths) and XML test results (JUnit/NUnit formats) are read from the filesystem into the agent context.
Boundary markers: Absent. The instructions do not define delimiters or specific safety warnings to prevent the agent from following instructions potentially embedded in log content.
Capability inventory: The skill utilizes Write, Edit, and Bash tools to modify source code (e.g., adding skip annotations) and create reports based on parsed log data.
Sanitization: Absent. No explicit validation or escaping is applied to the content extracted from logs before it influences code edits or command parameters.

test-flakiness