self-improvelisted

**Version**: 1.0 **Created**: 2026-05-11 **Last Updated**: 2026-05-11 **Authors:** Ömer Ufuk --- # /self-improve — Self-Improvement Cycle (L1 propose-to-file) **Usage**: `/self-improve` or `/self-improve --dry-run` or `/self-improve --max-goals N` > **Note**: Permitted under D-010's freeze per D-014 (Phase 0/1 self-improvement infrastructure is ecosystem-maintenance; scheduled automation waits until 2026-06-27). > **Merge dependencies**: this PR must merge AFTER #129 (ADRs D-014/D-015), #130 (autonomy rules), and #131 (baselines) — it references them. --- ## What This Command Does This is the **entry point** of the-matrix's autonomous self-improvement loop. It consumes the `dispatch_signals[]` produced by the `/self-state` perception layer and turns the top-ranked ones into structured improvement **proposals** the founder reviews and hands to the manager pipeline manually. It operates at **L1 (Suggestive)** — the agent proposes; the human executes. Per the founder's "safer first version" decision (2026-05-11), this loop: - Writes proposals to files. It does **not** change code, templates, or config. - Does **not** create git branches. - Does **not** open pull requests. - Does **not** dispatch the manager autonomously. - Does **not** merge anything. The founder reads each proposal and, if they choose to act, runs the manager manually using the exact invocation embedded in the proposal. --- ## Why L1 The full architecture (D-014) and the autonomy-level ADR (D-015)