Added /Users/samaguiar/Documents/Projects/routines-build/scripts/skill_surface_check.py, a read-only checker that reports skill and agent-instruction coverage across true skill loaders, editor/MCP surfaces, and API/model surfaces. It covers Codex, Claude, Cursor, VS Code, ChatGPT/OpenAI, Perplexity Computer/API, Gemini, Kimi, OpenRouter, Manus, and other manifest-backed tools.
Updated /Users/samaguiar/Documents/Projects/Skills/skills-scheduled-mirror-sync/SKILL.md so the nightly mirror sync appends this cross-tool surface check to its run report. Also added the same cross-tool step to the existing Claude Scheduled mirror copy at /Users/samaguiar/Documents/Claude/Scheduled/skills-scheduled-mirror-sync/SKILL.md so the scheduled task can see the new instruction path.
Ran python3 routines-build/scripts/skill_surface_check.py --markdown: pass, 9 surfaces checked, 0 surfaces needing attention, 13 same-slug drift warnings. Ran python3 -m py_compile routines-build/scripts/skill_surface_check.py: pass. Ran python3 routines-build/scripts/skill_registry_check.py: existing registry validator still fails on 36 preexisting skill-index and non-skill-folder issues, separate from this change.
Task is functionally complete. The new checker is read-only and does not copy, delete, or rewrite skills. It intentionally treats API/model surfaces such as Gemini, Kimi, Perplexity, and ChatGPT as manifest, MCP, CLI, or connector covered unless a real local skill folder exists.
A future agent may want to triage the 13 same-slug drift warnings and the older SKILL-INDEX.md validator failures as a separate cleanup pass. It would also be useful to promote this to a cloud-visible endpoint once Sam approves cloud migration for local routines, so multiple machines and agents can read the same surface report.
Requested routine expansion was implemented and QA completed. Remaining work is optional cleanup, not required for the expanded sync/check routine to function.
Do not overwrite tool-specific skills just because the checker reports same-slug drift. Review each mirror first, because Cursor, Codex, Claude Scheduled, and Claude custom skills may intentionally differ from canonical Projects skills.