The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external build logs, creating a potential surface for indirect prompt injection.
Ingestion points: Build logs are fetched using ReadMcpResourceTool and curl based on build UUIDs.
Boundary markers: The instructions do not specify any delimiters or safety markers (e.g., XML tags or "ignore instructions" prompts) when the agent processes the log content to extract errors.
Capability inventory: The agent uses curl, the gh CLI, and specialized MCP tools for job/build searching and resource reading. It also performs file uploads for HTML delivery.
Sanitization: No explicit sanitization or filtering of the log content is mentioned before the agent analyzes it or interpolates it into the final report.
[COMMAND_EXECUTION]: The skill utilizes system commands to perform its workflow.
The gh CLI is used to list pull requests based on commit SHAs (gh pr list --search <commitSha>).
The curl utility is used to download build logs from presigned S3 URLs (curl -sL "{url}" | tail -50).
[EXTERNAL_DOWNLOADS]: The skill downloads content from external S3 URLs.
Presigned S3 URLs are retrieved via an MCP tool call (ReadMcpResourceTool) and subsequently accessed using curl to retrieve build logs for analysis.

ci-health-report