Sift
Evaluate an outside repo — one someone linked, starred, or suggested — before any of it gets adopted into this repo or a sibling, across three axes: redundancy against what's already built, planned, or previously rejected here and in Word_Burner / Covenant / vibecoding-common-sense; license and provenance obligations measured against the destination repo's own license and distribution model rathe…
Skill Observer
Notice and log when a skill in .claude/skills/ produces wrong or incomplete output, gets corrected, or gets manually worked around. Use whenever you are actively running one of your own skills and get corrected on its output, a documented step turns out stale or missing a case, or its instructions conflict with something CLAUDE.md now says. Not for general working-style feedback about the agent i…
Model Check
Recommends which model — a local model, Codex, or Claude Haiku / Sonnet / Opus — and, for Sonnet and Opus, which effort tier fits the task you're about to do, so you don't default to a high-effort model for simple work. Also decides when a task should be split across an orchestrator and cheaper worker subagents rather than run on one model. Use this whenever you run /model-check, ask "what model…
Recon
Copy a skill, hook, or tool from this repo into one or more sibling repos (Covenant, vibecoding-common-sense) and register it in docs/parity-manifest.json plus the cross-repo-parity workflow's trigger paths, so future drift gets caught by CI instead of going unnoticed the way model-check once did. Use when the user says "/recon", "share this skill with Covenant", "copy this across repos", "keep t…