The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform repository management tasks. It executes git for branch creation and commits, and gh (GitHub CLI) to submit optimization proposals as draft Pull Requests. This process is designed for administrative productivity within a development environment.
[EXTERNAL_DOWNLOADS]: The instructions reference specific, version-pinned MCP servers (awslabs.cloudwatch-mcp-server and awslabs.prometheus-mcp-server) provided by the awslabs organization on PyPI. These are recognized as legitimate resources from a well-known vendor.
[DATA_EXFILTRATION]: The skill accesses telemetry data, including production traces and logs from CloudWatch, Prometheus, and Langfuse. This data ingestion is the primary function of the skill to identify quality regressions and potential PII leaks; the data remains within the local development context and the specified project directories.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it processes untrusted production traces to generate prompt modifications.
Ingestion points: External production traces are fetched via the mcp__langfuse__query_traces tool.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the analyzed traces.
Capability inventory: The agent has access to the Bash tool and the ability to modify local files and create git commits and Pull Requests.
Sanitization: While no automated sanitization of trace content is specified, the skill is architected to submit all findings as draft Pull Requests, which mandates manual human review and approval before any changes are integrated.

self-improving-loop