Monte Carlo Instrument-Agent Skill

This skill walks an MC Agent Observability customer through instrumenting a new AI agent in their Python codebase: detect AI libraries → install the Monte Carlo OpenTelemetry SDK + matching instrumentors → generate mc.setup() (with SimpleSpanProcessor when serverless) → propose @trace_with_workflow / @trace_with_task decorator diffs → confirm env vars (only when needed) → verify traces flow via get_agent_metadata.

The skill produces traces. It is not for monitoring or alerting on existing traces — that's monte-carlo-monitoring-advisor. The two skills are sequential: instrument-agent first, monitoring-advisor afterward.

Reference files live next to this file. Use the Read tool (not MCP resources) to access them.

CRITICAL — Never modify any file without explicit user approval

This skill must not modify any file in the customer's codebase without explicit per-file user approval. This rule covers:

• Dependency files — requirements.txt, pyproject.toml, Pipfile, lockfiles. Always propose the diff and wait for confirmation before editing.
• Source code — mc.setup() insertion, decorator placement (@trace_with_workflow, @trace_with_task), import additions. Always propose the diff and wait for confirmation per file.
• Env files — .env, .envrc, shell rc files. Always propose the change and wait for confirmation before editing.

The skill walks the user through what needs to change and why, then proposes diffs. It does not apply edits, run pip install, or write env files autonomously. The only exception: the user may explicitly waive approval for a specific file ("I know the risks, just edit the file") — proceed for that file only and surface that the approval was waived.

This guardrail is reinforced in the Tier-3 references (references/decorator-placement.md, references/setup-template.md, references/library-detection.md).