The Agent Skills Directory

[SAFE]: The skill follows security best practices by explicitly instructing the agent never to execute the crashed program, even when requested for reproduction. This mitigates the risk of executing malicious or data-damaging binaries.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted binary data from core dumps and ELF files. While this represents a potential attack surface for indirect injection, the risk is low because the skill uses structural analysis tools (gdb, objdump, REA) in a read-only, non-execution mode and contains instructions to treat data as evidence rather than commands.\n- [COMMAND_EXECUTION]: The skill invokes standard system utilities (gdb, objdump, python3) for forensic analysis of local files. Command execution is constrained with restrictive flags (e.g., gdb's -nx and -batch) to prevent the loading of untrusted initialization files or execution of code.\n- [DATA_EXPOSURE]: The instructions provide explicit privacy warnings regarding the nature of core dumps, which can contain process memory with credentials or PII, and strictly forbid the exfiltration or uploading of such data.

crash-triage