Shared memory and context tools for agentic work.
Code Rooms
# v1tals — m1nd — 2026-07-23 (bulletin 3)
> Triggered by the 5-merges standard (hook rang at 10 merges since `docs/vitals/2026-07-21b.md`).
> The window IS the 1.5.0 release era: MCP schema contract, honest recovery playbook, legacy
> snapshot adoption, the partial Windows heal + phase-2 declaration, five release-ceremony fixes,
> and the owner-ratified PRD re-freeze. 43 files, +1477/−198 — **zero UI-facing changes**, so the
> live UI axes (design/ergonomics/animations) carry over from bulletin 1 unchanged and were not
> re-driven. Scope: `main` @ `68545658`.
## Trend vs bulletin 2 (2026-07-21b)
| Eixo | Nota | Trend | Evidência |
|---|---|---|---|
| 0 · Era-coerência dos docs | **A−** | up | README versions, CHANGELOG [1.5.0], PATHOS cp31+completion, HANDOFF re-freeze note, AGENTS.md (+2 contract laws), skills — all updated in-era. Two gaps found THIS checkup, both fixed inline: the CI-contract test still pinned the pre-`./` npm publish string (main CI red — healed here), and IDE-INTEGRATIONS promised a Windows `.exe` the release doesn't ship (now states phase-2 + build-from-source). |
| 1 · Dívida (backend) | **C+** | flat (good) | Monoliths held through a release era: `layer_handlers.rs` 15,814 (=), `server.rs` 14,138 (+181, schema fixes+tests), `external_mutation_service.rs` 11,135 (+18), `enclave_authority.rs` 2,089 (below the 2,500 watch line). New module `legacy_snapshot_adoption.rs` 416 (cohesive, journaled). **Zero TODO/FIXME/HACK** across all four crates — held through +1,477 lines. |
| 2 · E2E real | **A** | up | The most adversarial e2e possible today: **installed the PUBLISHED 1.5.0 as a stranger** — `npm pack @maxkle1nz/m1nd@1.5.0` resolves and unpacks (CLI complete); `cargo add m1nd-mcp@1.5.0` resolves and downloads from crates.io in a clean sandbox. The world really can install what we shipped. |
| 3 · Design | **A−** | held | No UI delta this window; bulletin-1 verdicts stand. |
| 4 · Ergonomia | **A−** | up | Two honest-surface wins shipped: empty-brain recovery playbook now tells the truth (policy-aware steps instead of a refusal loop), and win32 `m1nd update` explains phase-2 instead of an obscure ENOENT. |
| 5 · Animações | — | held | No UI delta; not re-driven. |
| 6 · Performance | **A** | held | New binary boot: graph load 5,540 nodes < 1s in probes; served-owner latencies from bulletin 1 stand. Release-gate full suite is ~65min in CI — flagged as the CI-tiering front, not a runtime regression. |
## Fixes dispatched (the checkup is not a drawer report)
- **Inline (this bulletin's commit):** CI-contract test updated to the `./`-prefixed publish
string (heals the only red job on main); IDE-INTEGRATIONS Windows sections rewritten to
phase-2 truth.
- **Already in flight from the era:** Windows source-edit path-canon suite (single diagnosed
root cause, GOD-loop follow-up), updater cosign trusted-path product bug, brain_runtime
ReadDir sibling (task chip filed).
## New watch items
- `server.rs` crossed 14k lines (+181). Still cohesive (registry+handlers+tests), but it is
now the #2 monolith; revisit at ~15k or at the next architecture verdict alongside
`layer_handlers`.
- Release-gate wall-clock (~65min) makes every tag ceremony expensive — the CI-tiering front
(fast-on-push / heavy-on-nightly+release) is the named cure, queued.
## Standing debt (unchanged)
`layer_handlers.rs` 15,814 remains #1, verdict-gated before any costly refactor — the release
era did not worsen it.
## Next bulletin
- Re-drive UI axes on the first UI-facing change (none since 2026-07-21).
- Check `server.rs` against the 15k line and `enclave_authority.rs` against 2,500.
- If the genesis wave lands: e2e the governed bootstrap as the new critical flow.
