docs(agent): stability+usefulness survey #71 — 2 PROPOSED items

Research iteration (LOOP §1). Findings: the state-file path is robust;
the real stability hole is the eval-only automated lock-bump gate (#14)
with no binary cache, whose true fix is the KVM-blocked #20. Usefulness:
the capture suite lacks a screenshot annotate step. Filed 2 PROPOSED —
an interim build-tier lock gate (no KVM) and a satty annotate step.
V0 (docs-only).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

agent/BACKLOG.md

@@ -324,6 +324,40 @@ implement. Bernardo moves accepted items into a tier.*
   pairs with a hardware-queue session. Recommend (a) now, (b)
   investigated when the T14s is next available.
 
+- **[stability] Interim build-tier gate for automated lock bumps**
+  (surfaced by the 2026-07-08 stability survey, iteration #71). Today
+  `.gitea/workflows/bump.yml` pushes a lock bump to `main` on the
+  **eval tier only** (`nix flake check --no-build`), and Nomarchy ships
+  **no binary cache** — so users pull `main` and *compile* whatever
+  landed. A bump that evaluates but fails to *build* (a dropped patch, an
+  IFD/Stylix break, an upstream package that no longer compiles) reaches
+  users as a broken `home-update`/`sys-rebuild`. Item 20 (VM suite in CI)
+  is the real fix but is `[human]`-blocked on a KVM runner. Cheap interim
+  that needs **no KVM**: have the bump gate additionally *build*
+  `homeConfigurations.nomarchy.activationPackage` and
+  `nixosConfigurations.nomarchy.config.system.build.toplevel` before the
+  push (V1 tier — catches build-time breakage eval misses). Cost/open
+  question: the docker `act_runner` compiles Nomarchy's own derivations
+  from source (cache.nixos.org still serves upstream), so a full toplevel
+  build could be slow/heavy — measure once; if too heavy, gate on the HM
+  activation package alone (the surface most bumps touch). Does not
+  replace item 20; it raises the floor until item 20 lands.
+
+- **[usefulness] Screenshot annotation step (satty)** (surfaced by the
+  2026-07-08 usefulness survey, iteration #71). The capture suite is rich
+  (region/screen → clipboard or timestamped file, OCR, recording) but has
+  **no annotate step** — you can't arrow/box/blur/crop before sharing,
+  the single common workstation gap left. Add `satty` (Wayland-native,
+  themeable) and an "annotate" path: e.g. `grim -g "$(slurp)" - | satty
+  --filename -` with save-to-`~/Pictures/Screenshots` + copy-to-clipboard
+  actions, wired as a Print-family keybind (SUPER+SHIFT+Print, or a
+  Tools ▸ Capture ▸ "Annotate region" row) with the cheatsheet entry.
+  Cost: small — one package + one keybind/menu row + satty theming to the
+  palette (its config takes colors). Verifiable to V1 headlessly (build +
+  bind renders); the annotate UI itself is V3-visual (queue it). Check
+  whether satty's palette knobs are enough to keep it on-theme or if it
+  needs a generated config file like the other surfaces.
+
 ## Decisions `[human]`
 
 Open calls only Bernardo can make; agents add options/evidence but never

agent/JOURNAL.md

@@ -17,6 +17,26 @@ Template:
 
 ---
 
+## 2026-07-08 — Stability + usefulness survey → 2 PROPOSED (iteration #71)
+- **Task:** Bernardo asked to find the next tasks that make the distro more
+  stable + more useful. Research iteration (LOOP §1: write up PROPOSED).
+- **Did:** mapped the module/pkg surface + HARDWARE-QUEUE. Findings:
+  (stability) the state-file path is robust (friendly JSON errors, validate
+  cmd, atomic rebuild on corruption — not a gap); the real hole is the
+  automated lock-bump channel (#14) pushing to `main` on **eval-only** with
+  **no binary cache**, so untested runtime/build breakage compiles straight
+  onto users — the true fix is #20 (VM-suite-in-CI, `[human]`, KVM-runner-
+  blocked). (usefulness) the capture suite is rich but has no screenshot
+  **annotate** step (no satty/swappy); clipboard/emoji/colorpicker/OCR/
+  recording all already ship. Filed 2 PROPOSED: an interim build-tier lock
+  gate (no KVM needed, raises the floor until #20) and a satty annotate step.
+- **Verified:** V0 (docs-only; flake check was green in #70, tree unchanged).
+- **Pending:** Bernardo to triage the 2 PROPOSED into tiers, and — the
+  highest-leverage stability action — decide whether to unblock #20 (register
+  a KVM runner) so the lock-bump channel gets a real runtime gate.
+- **Next suggestion:** if #20 stays blocked, the interim build-tier gate is
+  the cheapest stability win; satty is the cheapest usefulness win.
+
 ## 2026-07-08 — QA sweep: clean tree, no actionable drift (iteration #70)
 - **Task:** QA sweep — headless-actionable queue is empty (NEXT is all
   `[human]`/`[blocked:hw]`/V3-visual: #29 hw, #28 icon-weight [human]+visual,