The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external diagnostic data such as error reports, logs, and stack traces via the $ARGUMENTS placeholder. This creates a surface for indirect prompt injection if the processed logs contain malicious instructions intended to influence the agent.
Ingestion points: The $ARGUMENTS variable in SKILL.md serves as the primary entry point for untrusted diagnostic data.
Boundary markers: The instructions do not define specific delimiters or unique markers to isolate the untrusted input from the agent's instructions.
Capability inventory: The skill instructs the agent to reproduce issues and perform experiments, which implies access to environment-specific tools and command execution.
Sanitization: The SKILL.md includes a safety instruction to "Redact secrets and PII from shared diagnostics," providing a level of protection against sensitive data exposure during analysis.
[COMMAND_EXECUTION]: The debugging workflow involves "Gathering context" and "Reproducing the issue with targeted experiments," which entails the execution of system commands, scripts, and profiling tools as described in the implementation playbook. These operations are restricted to the context of debugging the provided application code.
[EXTERNAL_DOWNLOADS]: The documentation and code samples provide instructions for integrating with well-known observability and monitoring platforms, including Sentry, DataDog, and OpenTelemetry. These references are consistent with the skill's primary purpose of implementing production-grade error handling and monitoring.
[DYNAMIC_EXECUTION]: The playbook contains code samples for runtime profiling and memory analysis, such as using Python's cProfile and Node.js's v8 heap snapshotting. These tools are used for diagnostic purposes on local application code during debugging sessions and do not involve loading code from untrusted remote sources.

error-debugging-error-analysis