Imported from caniko/fragpipe-mcp (
.skills/fix-loop-cross-platform-1v1/SKILL.md). Install upstream withnpx skills add caniko/fragpipe-mcp --skill fix-loop-cross-platform-1v1. Copyright stays with the author.
Cross-repository work: If scope spans repositories, invoke $graphify before discovery, planning, or edits. Query an existing graph; build/update a merged graph when missing, stale, or incomplete. Reuse a current graph for the same repository set.
Fragpipe Cross-Platform 1v1 Fix Loop
Use the fragpipe MCP server's repeat_cross_platform_1v1 tool.
Required parameters:
workdir: project checkout containing fragpipe configs.native_config: native WebRTC config, usuallyfragpipe.toml.android_config: Android config with test-peer and UI APK settings.max_runs: current fix-loop tier run count.timeout: default per-run timeout, usually300.
Run fail-fast first. If only one cell fails, switch to the focused WebRTC, Android WebRTC, or Android UI skill and fix that path before returning to the matrix.
When the Android UI cell captures a screenshot and the remaining question is
semantic visual quality rather than launch/logcat/dimension correctness, use
the global visual-rubric skill for screenshot rubric evaluation.