The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from production environments, creating a potential injection surface.
Ingestion points: The gcx agento11y experiments pull command downloads conversation payloads, trial inputs, and backend error text from production logs into the ./exports/ directory.
Boundary markers: The instructions explicitly warn the agent to treat all exported and derived data as untrusted data and never as instructions, though no technical delimiters are enforced on the content itself.
Capability inventory: The skill possesses Bash, Read, Write, and Edit capabilities, allowing it to execute CLI commands and modify files based on its analysis of the ingested data.
Sanitization: No automated sanitization is described; the skill relies on the agent's internal safety guardrails and the provided instructional warnings.
[COMMAND_EXECUTION]: The skill makes extensive use of the gcx CLI tool to manage Grafana resources, including creating, updating, and deleting evaluation rules and evaluators via the shell.
[EXTERNAL_DOWNLOADS]: The skill fetches observability data and experiment bundles from Grafana's backend infrastructure using the gcx utility as part of its intended diagnostic purpose.

agento11y