The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data, including pull request diffs, contributor design documents, and external web content for comparisons. This creates a potential surface for indirect prompt injection if malicious instructions are embedded in the analyzed materials.
Ingestion points: Workflow steps in SKILL.md involve acquiring target branches, PR text, and external documentation.
Capability inventory: The skill is authorized to write files to the local system (tmp/reports/ or report/) and uses an execution proxy for terminal commands (rtk).
Sanitization: The skill includes a 'Security & Multi-Agent Hygiene' section that mandates treating web content and external data as hostile and requires redacting secrets/PII before sharing context.
Boundary markers: The instructions emphasize 'think before coding' and procedural verification gates, though they do not specify unique string delimiters for the untrusted content itself.
[EXTERNAL_DOWNLOADS]: The HTML template in assets/html-report-starter.html fetches font resources from Google Fonts (fonts.googleapis.com and fonts.gstatic.com). These are well-known services used for visual styling and do not represent a security risk.

pr-report