The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is instructed to ingest and process untrusted data from external sources while possessing capabilities that could be abused if malicious instructions are embedded in that data.
Ingestion points: The agent is directed to read user-provided codebases, architecture documents, README files, and fetched Datadog dashboard definitions (via pup dashboards get) to gather context for dashboard design and auditing.
Boundary markers: The instructions lack requirements for using delimiters or "ignore embedded instructions" directives when handling this external content.
Capability inventory: The skill has access to Shell, Write, Read, and AskUserQuestion tools, providing a significant attack surface if an injection occurs.
Sanitization: There is no mention of sanitizing or validating the content read from files or CLI outputs.
[COMMAND_EXECUTION]: The skill relies on the pup CLI tool to interact with the Datadog environment. It executes multiple shell commands to list dashboards, retrieve definitions, and explore telemetry metrics.
Evidence: Usage of commands such as pup dashboards list --agent, pup dashboards get <id> --agent, and pup metrics list --filter="<service-name>.*" --agent is central to the skill's operational workflows.
[DYNAMIC_EXECUTION]: The skill provides a Python script template in references/tabs.md for auditing dashboard JSON for duplicate widget IDs. The agent is expected to use this logic to perform validation tasks during the audit process.
Evidence: The skill documentation includes a Python script utilizing json and collections.Counter and instructs the agent to "audit all widget IDs for duplicates" before modifying dashboard structures.

datadog-design-dashboard