The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides scripts (di_instrumentation.py, di_snapshots.py) that execute authenticated AWS API calls to manage instrumentation configurations and query CloudWatch Logs. These scripts are hardened against abuse through literal attribute access patterns that prevent arbitrary method invocation on the AWS clients and use structured JSON inputs to avoid shell injection vulnerabilities.- [EXTERNAL_DOWNLOADS]: The skill's guides instruct users to download and install instrumentation agents and SDKs (such as the AWS Distro for OpenTelemetry) from official AWS and OpenTelemetry repositories on GitHub and public package registries. These are standard, legitimate resources for enabling the observability features provided by the skill.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from running applications via CloudWatch Logs snapshots. This creates a surface for indirect prompt injection. The skill mitigates this risk by providing clear instructions for the agent to parse this data using programmatic tools like jq and Python instead of manual transcription, and by implementing strict validation for identifiers like location hashes.

aws-observability