crash-triage-reference
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Provides standard shell command templates (e.g.,
grep,sha1sum,tee) and fuzzer-specific flags (e.g.,-minimize_crash=1,-runs=1) to help users analyze local crash artifacts and deduplicate bug reports. - [INDIRECT_PROMPT_INJECTION]: The skill provides a framework for analyzing untrusted data in the form of fuzzer artifacts and sanitiser output reports.
- Ingestion points: Crash artifacts and sanitiser report files are the primary data sources for the triage workflow defined in SKILL.md.
- Boundary markers: The instructions rely on structured sanitiser output (e.g.,
==ERROR:,runtime error:) to delimit data, though no specific AI boundary markers are mentioned. - Capability inventory: The workflow involves reading files, executing shell pipelines for hashing, and running the target binary with specific artifacts.
- Sanitization: Analysis is performed on the raw output of sanitizers; the skill assumes the user/agent is performing a technical review of these reports.
Audit Metadata