ci-failure-triage
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could potentially contain malicious instructions aimed at influencing the agent's classification output.
- Ingestion points: The agent is instructed to read 'runner output', 'stack traces', 'run history', 'environment metadata', and 'diffs' (SKILL.md).
- Boundary markers: Absent. There are no instructions for the agent to use specific delimiters or to treat the ingested CI data as untrusted text strings rather than instructions.
- Capability inventory: The skill itself does not define tools, but the triage workflow assumes the capability to read files and potentially execute shell commands to gather signals.
- Sanitization: Absent. The instructions do not specify any filtering or escaping of content extracted from logs or code diffs before the agent processes them.
- Mitigation Note: The skill's requirement for a structured output that includes rejected alternatives and specific evidence for each signal significantly mitigates the risk of successful exploitation by requiring logical consistency.
Audit Metadata