The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates as a CLI wrapper for ae-cli, allowing the agent to manage experiments, features, and metrics through structured subcommands and JSON request payloads. These operations are project-scoped and follow an established administrative workflow.
[INDIRECT_PROMPT_INJECTION]: The skill utilizes a read-modify-write pattern that processes data from CLI query responses to populate parameters for future save commands, which introduces an indirect prompt injection surface.
Ingestion points: Data retrieved from the environment through commands like ae-cli experiment experiment get, ae-cli experiment metric list, ae-cli experiment feature list, and ae-cli experiment bucket list (found in references/query_experiment_detail.md, references/query_metric_list.md, etc.).
Boundary markers: The skill instructions do not prescribe the use of delimiters or specific warnings to ignore instructions that might be embedded in the data returned by the tool.
Capability inventory: The skill provides numerous high-privilege write capabilities, including ae-cli experiment experiment save, ae-cli experiment metric save, and ae-cli experiment feature save (found in references/save_experiment.md, references/save_metric.md, etc.).
Sanitization: There are no documented requirements for the agent to sanitize, validate, or escape the content retrieved from the CLI before using it in the construction of subsequent commands.

ae-experiment