ai-bug-triage
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines an automated pipeline that accepts external log inputs and automatically generates tracking issues, introducing an indirect prompt injection vulnerability surface.
- Ingestion points: Untrusted application errors, messages, and log excerpts enter the agent context via the LLM classification and ticket generation prompts specified in
references/pipeline-prompts-and-integration.md. - Boundary markers: Simple brackets (
[MESSAGE],[LOG EXCERPT]) are used as placeholders within the prompt templates; however, rigorous boundary delimiters and explicit defensive instructions instructing the LLM to ignore embedded commands are absent. - Capability inventory: The pipeline is equipped with integration capabilities to execute commands or call APIs that write data externally, specifically using the GitHub CLI (
gh issue create) to create issues automatically as detailed inreferences/pipeline-prompts-and-integration.md. - Sanitization: Structural log normalization is implemented in
SKILL.mdto strip system artifacts (PIDs, UUIDs, timestamps), but linguistic filtering, escaping, or verification of the raw text messages to prevent command injection or instruction overrides is absent.
Audit Metadata