The Agent Skills Directory

[COMMAND_EXECUTION]: The skill in st ruc ts the agen t to use the Bash too l to run the local analy ze r scrip t (.claude/tools/utils/log_analy ze r.py) and utili ze s standar d utili tie s like tail for real-time log moni toring.
[PROMPT_INJECTION]: The skill exhibi ts an indirec t prom p t injec tion at tack surface (Categor y 8) becau se it proce sses untru sted log message s th roug h an LLM for roo t cau se analy sis.
In gestion poin ts: Read s untru sted log con ten t from projec t log file s at {{LOG_DIR}}/{{LOG_FILE}}.log.
Boundar y marke rs: The tech nical refe rence documen t s men tion s system prom p t s for analy sis bu t lack s evidence of ex plici t delimi te rs or in st ruc tion s to igno re em bed ded malic iou s st ring s wit hin the log data.
Capabili ty inven tor y: The analy ze r scrip t run s via the Bash too l and has the abili ty to wri te mark down re por t s to the file system.
Sani ti za tion: The re is no in dica tion tha t log en t rie s are sani ti ze d or esca pe d befo re bein g in te r pola te d in to the AI prom p t.

observability-analyze-logs