The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script check_changelog.py performs a network request to raw.githubusercontent.com to retrieve the official VictoriaMetrics changelog.
Evidence: It fetches the public CHANGELOG.md files directly from the official VictoriaMetrics/VictoriaMetrics repository.
Context: This is standard functionality designed to assist users by highlighting performance fixes or features introduced in versions newer than the one analyzed in the trace.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external, potentially untrusted user-supplied JSON trace files, presenting an indirect prompt injection surface.
Ingestion points: The user-supplied <trace_file> is loaded and parsed by scripts/parse_trace.py.
Boundary markers: There are no explicit delimiters or strict prompt boundaries surrounding the parsed message text when reviewed by the model.
Capability inventory: The allowed tools are restricted to Read and Bash (specifically scoped to python3). There are no capabilities to write arbitrary files or transmit data to non-whitelisted external servers.
Sanitization: The parse_trace.py script clips long message fields using node["message"][:200], limiting exposure to oversized payloads. Furthermore, disable-model-invocation: true configuration protects the execution context from model-driven tool abuse.

vm-trace-analyzer