Shared memory and context tools for agentic work.
Code Rooms
# v1tals — m1nd — 2026-07-21 (bulletin 2)
> Triggered by the 5-merges standard (the post-merge hook fired at 5 merges since
> `docs/vitals/2026-07-21.md`). Delta bulletin: the 5 merges were the security era
> (candidate-source guard advisories fix, G4 phase-2 Windows, the Secure Enclave custody floor
> FU1+FU2). All backend/authority — **no UI-facing change**, so the first bulletin's live UI axes
> (646/646 unit, 31/31 e2e, north 282ms/seek 27ms) stand and were not re-driven. Scope:
> `main` @ `2bf9f25f`.
## Trend vs bulletin 1 (2026-07-21)
| Eixo | Nota | Trend | Evidência |
|---|---|---|---|
| 1 · Dívida (backend) | **C** | **flat** (good) | Monoliths did NOT grow through a ~2000-line authority era: `layer_handlers.rs` 15,814 (=), `server.rs` 13,957 (=), `external_mutation_service.rs` 11,117 (+7), `autonomy.rs` 5,371 (+39 custody_floor). The era added a **new cohesive module** (`enclave_authority.rs`, 2,073) rather than bloating existing monoliths. |
| 1 · Higiene | **A+** | **held** | Still **zero** TODO/FIXME/HACK across the whole codebase — through ~2,000 new lines of Secure Enclave / authority code. Remarkable discipline maintained. |
| 2 · Tests | **A** | **up** | The era added tests, didn't shed them: enclave 13 (incl. the persistence round-trip), release/autonomy custody-floor negatives, the triple-mirror smuggling tests. `m1nd-control` 137, `m1nd-mcp` lib 1419+. (Note: bulletin 1's "244 test files" was repo-wide; not directly comparable to a narrower scope — the honest signal is coverage grew.) |
| 4 · Ergonomia / 5 stati | **A−** | held | No UI change this window; the tested-invariant culture stands. |
| 6 · Performance | **A** | held | No hot-path change; north/seek latency from bulletin 1 stands. |
## New watch item
`enclave_authority.rs` at 2,073 lines is now a substantial module. It is **cohesive** (one Secure
Enclave adapter, cfg-macOS, 13 tests, zero debt markers) — not a grab-bag, so not debt today. But
it is the era's largest new surface; if it keeps growing past ~2,500 lines or accretes unrelated
responsibilities, revisit for a split. Watch, don't act.
## Standing debt (unchanged from bulletin 1)
`m1nd-mcp/src/layer_handlers.rs` (15,814 lines) remains the #1 backend debt, unchanged. Still
deferred to an askGOD verdict before a costly authority-plane refactor — not touched blind. The
security era did not worsen it, which is the right outcome for a feature-adding stretch.
## Fixes dispatched
- None required this bulletin — the trend is flat-to-improving. The standing `layer_handlers`
refactor remains a verdict-gated future item, not dispatched.
## Next bulletin
- Re-drive the UI axes (browser + design pass) once a UI-facing change lands.
- Re-check the `enclave_authority.rs` line count against the ~2,500 watch line.
- The `layer_handlers` refactor: is it time to bring the verdict?
