The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection vulnerability surface because it processes external, untrusted content from GitHub resources without explicit containment or validation rules.\n
Ingestion points: Outlined in references/workflow.md, data streams are fetched via GitHub search tools capturing issue titles, comments, PR descriptions, and active discussion threads.\n
Boundary markers: The skill instructions lack explicit boundary markers or delimiters to wall off external content when generating report markdown or HTML templates.\n
Capability inventory: The skill writes report files to the workspace (.github/reviews/briefings/) and invokes a localized rendering script via Node.js.\n
Sanitization: No verification or character escaping mechanisms are implemented for the aggregated third-party descriptions before embedding them into markdown/HTML outputs.\n- [COMMAND_EXECUTION]: The skill invokes local command execution to render the final briefing artifacts.\n
Evidence: The output contract in SKILL.md mandates running the command node skills/a11y-core/scripts/render-report.mjs over the collected workspace findings.

daily-briefing