The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and displays external telemetry data (OpenTelemetry spans) retrieved from PostHog which could theoretically contain adversarial content.
Ingestion points: The scripts print_summary.py, print_timeline.py, extract_span.py, and search_spans.py read JSON data files generated by PostHog tools.
Boundary markers: There are no explicit instructions or delimiters used to warn the agent to ignore embedded instructions within span names, attribute values, or error messages.
Capability inventory: The skill is restricted to reading, parsing, and printing the trace data; it does not perform network exfiltration or filesystem writes.
Sanitization: Trace attributes and operation names are displayed directly as text, providing a potential surface for injection if the data contains malicious payloads.
[COMMAND_EXECUTION]: The skill workflow involves executing local Python scripts included in its package to analyze data.
Evidence: The skill instructs the agent to run python3 commands on scripts such as scripts/print_summary.py using persisted JSON files as input.
Input parameters like SPAN and SEARCH are passed to these scripts via environment variables, which the agent sets based on user queries.

exploring-apm-traces