observability

Trace cross-agent LLM and tool interactions with append-only JSONL spans.

Updated Apr 21, 2026
One-click install
npx skills add https://github.com/brucebanner010198-commits/DevSecOps-Agency --skill observability-brucebanner010198-commits
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: observability
Source: https://github.com/brucebanner010198-commits/DevSecOps-Agency/tree/main/skills/observability
Command: npx skills add https://github.com/brucebanner010198-commits/DevSecOps-Agency --skill observability-brucebanner010198-commits

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Establishes a cross-agent tracing framework using OTel GenAI conventions to provide end-to-end visibility across CEO, Chiefs, specialists, and LLM/tool/retrieval calls, while enforcing data-safety boundaries.

Core Features & Use Cases

  • Spans and events: Each root and nested dispatch opens spans; per-agent calls, tool calls, and retrieval steps create child and leaf spans. Spans are stored append-only as _vision/traces/<yyyy-mm>/<trace-id>.jsonl and referenced via traceId.
  • Attributes and events: Follows gen_ai.* attributes for system, request model, operation name, token usage, and tool data; prompts/completions are stored as span events rather than attributes.
  • Safety and observability: Tail-based sampling, data leakage prevention, and strict hygiene rules prevent prompts/completions from entering attributes; spans provide an auditable cross-agent view.
  • Use Case: Debugging complex multi-agent interactions, capacity planning, and security reviews across CEO, Chiefs, and specialists.

Quick Start

Initiate cross-agent tracing by starting a root CEO span for the next dispatch and ensure all subsequent LLM/tool/retrieval calls create child spans under the same trace.

Frequently Asked Questions about observability

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How does cross-agent tracing work for multi-agent LLM interactions?

Cross-agent tracing opens spans for each root and nested dispatch across CEO, Chiefs, and specialists, storing append-only child spans in JSONL files. This provides end-to-end visibility across LLM, tool, and retrieval calls referenced via traceId.

How do I prevent prompt data leakage in distributed tracing telemetry?

To prevent prompt data leakage in distributed tracing, data leakage prevention and strict hygiene rules ensure prompts and completions are stored as span events rather than attributes, keeping sensitive content out of standard telemetry fields.

What is tail-based sampling in OTel GenAI conventions?

Tail-based sampling in OTel GenAI conventions evaluates complete traces before deciding retention, ensuring spans with gen_ai.* attributes for system, model, and token usage are safely sampled for auditable cross-agent telemetry.

How do I debug complex multi-agent interactions across LLM and tool calls?

Debug complex multi-agent interactions by initiating a root CEO span for the next dispatch and ensuring all subsequent LLM, tool, and retrieval calls create child spans under the same traceId for a complete auditable view.

Where are distributed tracing span events stored for security auditing?

Distributed tracing span events are stored append-only in _vision/traces/<yyyy-mm>/<trace-id>.jsonl files. This cross-referenced JSONL format enforces data hygiene and ensures safe, auditable telemetry for security reviews.

Can I use OTel GenAI attributes for capacity planning in multi-agent systems?

Yes, OTel GenAI attributes capture system, request model, operation name, and token usage data across multi-agent spans, providing the auditable cross-agent view needed for effective capacity planning.