The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface by instructing the agent to read and process untrusted external files from the workspace repository.
Ingestion points: The agent parses local project files, including README.md, application entry points (main.*, index.*), and configuration files (meta.yaml, values.yaml, Deployment, Dockerfile) in SKILL.md (Phase 1) to determine the application's telemetry structure.
Boundary markers: The instructions lack specific delimiters or boundary markers to encapsulate the contents of these files or explicit safety guidance to ignore any instructions embedded within them.
Capability inventory: The skill utilizes powerful local capabilities, including executing numerous cx CLI subcommands (cx metrics, cx logs, cx spans, cx dashboards) to query telemetry and deploy configurations, as well as creating local dashboard files in /tmp/ as documented across SKILL.md and references/deploy.md.
Sanitization: No input sanitization or validation protocols are specified for terms or instructions extracted from the repository files prior to their inclusion in commands or prompt interpolation.

cx-dashboards