flaky-test-diagnosis

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests test failure output and signals, which are potentially attacker-controlled (e.g., in a CI/CD environment or through contributed code). This data is used to inform the agent's diagnosis and reproduction steps.
  • Ingestion points: The argument-hint in SKILL.md specifies that the agent receives "failure output" as input to the skill.
  • Boundary markers: The instructions lack specific boundary markers or warnings to ignore potential instructions embedded within the failure logs.
  • Capability inventory: The agent is instructed to execute build/run tooling (e.g., ctest, setarch) and inspect source code based on its analysis of the ingested data.
  • Sanitization: No sanitization or validation of the failure output is required by the instructions.
  • [COMMAND_EXECUTION]: The skill explicitly directs the agent to execute shell commands to reproduce test failures, including ctest with various flags (--repeat, --gtest_shuffle, -j) and setarch -R to disable ASLR for ThreadSanitizer debugging. While these are legitimate developer tools, they represent a significant capability that could be misused if the agent is influenced by indirect prompt injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 03:41 PM