The Agent Skills Directory

[COMMAND_EXECUTION]: The skill enables the agent to write analysis reports to the local file system (e.g., in the 'evals/reports/' directory) if the user provides the '--output file' flag. This grants the agent a controlled file-write capability.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to the nature of LLM experiment analysis. 1. Ingestion points: Untrusted data consisting of LLM experiment inputs, outputs, and expected results are retrieved from Datadog via the 'mcp__datadog-llmo-mcp__list_llmobs_experiment_events' and 'mcp__datadog-llmo-mcp__get_llmobs_experiment_event' tools. 2. Boundary markers: The instructions lack explicit delimiters or 'ignore embedded instruction' warnings when processing these logs. 3. Capability inventory: The skill has the capability to write local files and create Datadog notebooks using the 'mcp__datadog-mcp-core__create_datadog_notebook' tool. 4. Sanitization: While the skill includes a strong instruction to mask or redact PII, it does not specify generic content escaping for the retrieved LLM logs to prevent the execution of instructions embedded within them.

experiment-analyzer