| name | source-command-repeatability-spine |
| description | Run repeatability-spine when the user cannot repeat the magic, a revision got worse, a route chose the wrong workflow, a patch introduced a regression, or a good output needs a preservation lock before revision. |
source-command-repeatability-spine
Use this skill when the user asks to run repeatability-spine, says a revision failed, says the rewrite got worse or flat, says the system cannot repeat the magic, says the route picked the wrong workflow, or wants to preserve what made a previous output work.
Operator Core Alignment
This project wrapper follows .agent/workflows/repeatability-spine.md as the
canonical behavior source. It must stay a thin compatibility wrapper and
preserve:
- preserve the good example before repair
- every run needs grounded evidence, one primary failure class, Preservation Lock, repair route, validation, regression guard, and replay prompt
- inaccessible conversations are pending evidence, not invented findings
- routing failures require a verifier query or routing feedback log before being called repaired
- mutation-capable repair routes require a Goal Packet before edits
- global
~/.codex behavior changes require workspace proof and explicit approval
- broad broken-harness triage routes to
/autopilot or /system-audit
- real Codex subagents require explicit authorization
- no competing behavior contract
Command Template
Read and execute the workflow at .agent/workflows/repeatability-spine.md.