The Agent Skills Directory

[PROMPT_INJECTION]: The skill captures stdout and stderr from test executions and incorporates them into a report for the agent to analyze. This creates a surface for indirect prompt injection where a test could produce output designed to trick the agent into ignoring instructions or performing unintended actions.
Ingestion points: Test output is captured by scripts/diagnose-flaky-test.sh into temporary files and then appended to flaky-test-diagnosis.md.
Boundary markers: The test output is delimited within the report using markdown triple backticks (```).
Capability inventory: The skill allows the agent to execute shell commands (specifically test runners like vitest, jest, playwright, and cypress) and read/write files within the project directory.
Sanitization: The captured test output is included raw; the script does not escape characters that could break out of the markdown code block or perform content filtering.
[SAFE]: The bash script diagnose-flaky-test.sh correctly uses array-based execution for shell commands, which prevents command injection even if test names or file paths contain shell metacharacters.
[SAFE]: Input validation is performed on the test file path (checking for existence and preventing path traversal) and the number of runs (restricted to an integer between 1 and 100).

flaky-test-diagnostician