The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from external sources including Grafana SLO definitions and GitHub runbooks. This presents a surface for indirect prompt injection if the external data contains instructions designed to manipulate the agent.
Ingestion points: Runbook content is fetched via the GitHub API (gh api) in SKILL.md; SLO definitions are retrieved via the Grafana Cloud CLI (gcx).
Boundary markers: The instructions do not specify the use of delimiters or clear separation between system instructions and untrusted external content.
Capability inventory: The skill is allowed to use the Bash tool, which permits execution of shell commands like gcx and gh.
Sanitization: There is no explicit requirement for the agent to sanitize or validate the fetched content before displaying excerpts or using it for diagnostic summaries.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation from GitHub to provide runbook excerpts for troubleshooting.
Evidence: It uses the gh api tool to fetch contents from repositories defined in the SLO metadata as seen in Step 6 of SKILL.md.
Context: The downloads target well-known services (GitHub) and are consistent with the skill's primary purpose of diagnosing infrastructure alerts.
[COMMAND_EXECUTION]: The skill relies on several command-line utilities to perform its diagnostic workflow.
Evidence: It uses gcx for Grafana operations, gh for GitHub interactions, and jq for JSON processing throughout SKILL.md.
Context: All commands are scoped within the legitimate operations needed to investigate SLO breaches as described in the skill metadata.

slo-investigate