What problem does it solve?
The Local Self-Evolution Loop's PROPOSE-to-APPLY seam was dead in production: the frozen Go applier never ran and its write-flag stays disabled. This Skill closes that gap by safely playing back already-approved decision.json records into the repository, writing only to designated evolvable surfaces while hard-rejecting any change targeting frozen doctrine files.
Core Features & Use Cases
- Frozen-allowlist enforcement: Validates every proposal target against a frozen allowlist, refusing writes to protected paths (including path-traversal and divergent-patch bypass attempts) and logging each refusal.
- Execution-meta forced gate: Mechanically refuses changes to applier/curator skill bodies, hooks, or hook-registration settings unless the decision carries a synchronous-approval marker from the orchestrator's user gate.
- Playback with audit trail: Applies the approved diff.patch via git apply, appends an apply-ledger.jsonl row, and commits one lsel-<proposal-id>-tagged Conventional Commit on the feature branch.
- Post-apply verification: Runs the proposal's verify_command with a timeout-retry-once policy, auto-reverts on a second failure, and requires a mandatory lint/format/type/test gate as an independent check.
- Use Case: After the curator approves a proposal to update a memory feedback file, invoke this Skill to validate the target, apply the patch, commit it with an lsel tag, and verify the change — automatically reverting if verification fails twice.
Quick Start
Ask the agent to drive an LSEL APPLY pass for the approved proposal in .moai/state/lsel/proposals, validating it against the frozen allowlist and committing it through lsel-apply.sh.
