The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze LLM trace data from PostHog, which naturally includes untrusted content such as raw user messages and prior model outputs. This creates a surface where instructions embedded in the data could potentially influence the agent's behavior during the debugging process.
Ingestion points: Trace content returned by posthog:query-llm-trace and posthog:query-llm-traces-list and subsequently processed by scripts in the scripts/ directory.
Boundary markers: The instructions do not provide explicit boundary markers or directives for the agent to ignore instructions contained within the trace data objects.
Capability inventory: The skill uses posthog:execute-sql and trace retrieval tools, and requires the capability to execute local Python scripts to parse data.
Sanitization: No sanitization is performed on the natural language content within the traces; scripts implement data formatting and length-based truncation only.
[COMMAND_EXECUTION]: The skill includes and instructs the agent to execute several Python scripts (scripts/extract_conversation.py, scripts/extract_span.py, scripts/print_summary.py, scripts/print_timeline.py, scripts/search_traces.py, scripts/show_structure.py) to process trace data. While the scripts themselves are part of the skill package and perform analytical tasks, they establish a local code execution surface.

exploring-llm-traces