TelemetryShield — Stop AI Deleting Your Logs

💡 Built from Real Dev Pain

This skill wasn't generated in a vacuum or built for hypothetical scenarios. As a developer actively building modular architectures, I repeatedly suffered from AI agents blindly breaking repository states, wiping out production telemetry, or corrupting directory paths. Every single skill in this laboratory was created out of sheer frustration to solve my own production bottlenecks. Battle-Tested via Live Runs: I have spent dozens of hours running these guardrails through aggressive live-session testing in active workspaces. Every instruction has been iteratively adjusted, logged, and hard-refined. This product was only published to the marketplace after I personally verified that it consistently stops the AI from hallucinating and safely protects the codebase in real-world scenarios. This skill blocks the agent from dropping telemetry until it proves nothing was lost, ensuring your production observability remains intact after every automated edit. Security & Execution

• Zero External Dependencies: The internal check engine runs locally using the Python Standard Library (stdlib).
• 100% Local & Private: No network connections are made. Your source code, trace IDs, and architecture maps never leave your machine.
• Safe Verification Gate: It functions as a programmatic, non-destructive audit layer that blocks the agent from finalizing file writes if a telemetry drop is detected.

What it does

• Telemetry Manifesting builds a comprehensive map of all log calls, tracing spans, and metric increments before any code is touched.
• Boundary Preservation wraps business logic updates around existing instrumentation rather than restructuring or removing them.
• Regression Verification compares the final patch against the initial manifest to detect and block any accidental deletions.
• Schema Enforcement ensures new or modified logic stays aligned with project-specific semantic conventions and logging keys.
• Explicit Authorization blocks the removal of any instrumentation unless the user provides a direct, specific instruction to do so.

How it works

1. Inventory: The agent scans the target file to list every tracer, logger, and counter instance, storing their exact locations.
2. Execute: The agent performs the requested refactor while treating the telemetry lines as immutable anchor points.
3. Audit: A post-patch comparison runs to verify that every item from the initial inventory is present in the new code.
4. Report: The agent outputs a verification status alongside the code, listing exactly what was preserved.

Frameworks & tools Works with OpenTelemetry (OTel), Prometheus, Datadog, Sentry, and standard logging libraries across Python, JavaScript/TypeScript, Go, and Java. Why this beats prompting it yourself Generic prompts like "don't remove logs" are frequently ignored by LLMs during complex logic shifts or heavy refactoring due to context window drift. TelemetryShield implements a strict manifest-and-verify loop at the schema level. If the AI agent accidentally drops a single span or counter, the runtime validation gate fails, forcing the agent to self-correct and restore the missing instrumentation before the code hits your IDE. Use cases

• Refactoring legacy middleware while keeping distributed tracing headers intact.
• Optimizing high-throughput loops without losing per-iteration metric increments.
• Transitioning from monolith to microservices where context propagation is critical.
• Updating error handling logic without losing existing logger.error call sites.

Known limitations Preserves existing instrumentation. Does not add new telemetry where none existed — pair with a metrics-design skill for greenfield observability. Does not fix broken schemas; its sole purpose is to block silent deletion. Requires read access to source files to build the manifest.