The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of processing untrusted data from agent traces.
Ingestion points: The evaluation workflow described in references/evals.md reads conversation history (user and assistant turns) directly from CloudWatch spans and traces.
Boundary markers: The provided configuration templates and CLI commands do not implement specific delimiters or instructions to ignore embedded prompts within the trace data being evaluated.
Capability inventory: The skill uses the Bash tool to execute agentcore and aws CLI commands which interact with the evaluation engine and cloud infrastructure.
Sanitization: There is no evidence of sanitization or filtering of trace content before it is processed by LLM-as-a-judge evaluators.
[COMMAND_EXECUTION]: The skill leverages the Bash tool to perform project management and environment configuration.
It executes the agentcore CLI to run evaluations, deploy infrastructure, and manage online monitoring configurations.
It provides and executes AWS CLI command templates (e.g., aws logs put-retention-policy, aws xray update-trace-segment-destination) to manage logging and tracing settings.
These operations are standard for the skill's primary purpose of optimizing and observing an AgentCore runtime environment.

agents-optimize