diff --git a/CLAUDE.md b/CLAUDE.md
index 83044dd6..6e0dd517 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -66,14 +66,38 @@ Commit freely (per task + the spec/plan docs). Never push without asking.
 
 ## Building & Testing
 
-`dotnet build ClaudeDo.slnx` requires .NET 9; on .NET 8 build individual projects with `-c Release` (a running Worker locks the `Debug` output).
+**Build the whole solution, test only what you touched.** Always `-c Release` — a running Worker locks the `Debug` output.
 
 ```bash
-dotnet build src/ClaudeDo.App/ClaudeDo.App.csproj -c Release      # pulls in Ui + Data
-dotnet build src/ClaudeDo.Worker/ClaudeDo.Worker.csproj -c Release
-dotnet test tests/ClaudeDo.Worker.Tests/ClaudeDo.Worker.Tests.csproj -c Release   # also: Data.Tests, Ui.Tests, Localization.Tests, Installer.Tests, Releases.Tests
+dotnet build ClaudeDo.slnx -c Release                                    # ~40s, catches cross-project breaks
+dotnet test tests/ClaudeDo.Worker.Tests/ClaudeDo.Worker.Tests.csproj -c Release --no-build \
+  --filter "FullyQualifiedName~Handoff"                                  # seconds, not minutes
 ```
 
+`.slnx` needs SDK 9+ (this machine is on 10). On an older SDK build the projects individually:
+`src/ClaudeDo.App` (pulls in Ui + Data) and `src/ClaudeDo.Worker`.
+
+### Do NOT run the full test suite to finish a task
+
+Measured 2026-09-02, sequentially after one build: **Worker.Tests 507s**, Ui.Tests 33s, Data.Tests 8s,
+Localization/Installer/Releases 3s each, build 42s — ~10 min for all six. Run in parallel across
+projects it is *worse*, not better (Worker.Tests 801s, Ui.Tests 82s): these tests create real git
+worktrees and spawn processes, so they thrash rather than share.
+
+With up to 5 implementer tasks running at once, a full suite per task costs hours — and **cannot find
+what it costs that for.** Each task works in its own worktree containing only its own change, so a
+collision between two parallel tasks is invisible to both. Collisions surface after the merge, which
+is exactly where `verify_merges` runs the list's verify command **once per batch**. The full suite is
+the merge gate's job, not the task's.
+
+So, per task:
+- **Build:** the whole solution, always. It's cheap and it catches the compile break your change
+  caused three projects over.
+- **Tests:** `--no-build --filter` scoped to what you touched. Add a filter for the classes covering
+  your change; if you genuinely can't scope it, run the one relevant test project, never all six.
+- **Report** what you ran and what you did not, so the merge session knows what is still unverified.
+- Never report "all tests pass" when you ran a filtered subset. Say which filter you used.
+
 ### Gotchas
 - **Subagents:** use the `sonnet` model; stage files explicitly by path — never `git add -A` (parallel sessions often leave unrelated WIP in the tree).
 - **Icons:** `PathIcon` *fills* its geometry — line-art icons must be filled geometry or a `Path Classes="icon-stroke"`, else they render invisible. Colour comes from the `ICON PAINTING` block in `IslandStyles.axaml` (class + state); setting `Foreground`/`Stroke` inline kills the hover state.
diff --git a/docs/explore-notes/conpty-sessions.md b/docs/explore-notes/conpty-sessions.md
index cc500e55..e4251c70 100644
--- a/docs/explore-notes/conpty-sessions.md
+++ b/docs/explore-notes/conpty-sessions.md
@@ -115,6 +115,22 @@ share the same prompt/model as their non-final counterpart and differ only in on
 rendered into the handoff kickoff file (`PromptKind.MergeHelperHandoff`) marking the final round
 and forbidding further reruns.
 
+The chain length is bounded server-side by `HandoffRoundTracker`, whose cap comes from
+`AppSettingsEntity.ListHandlerMaxHandoffs` (default 4 = two merge rounds; `HandoffMcpTools` reads it
+per call and `Register` clamps to min 1, so 0 can never mean "unbounded"). Past the cap a
+`wait`/`merge` request is silently coerced to its `_final` counterpart — check the result's
+`coerced` flag — and any handoff after a `merge_final` round throws.
+
+**PowerUse.** When `AppSettingsEntity.PowerUseEnabled` (toggled on the usage monitor) **and** the
+list's own `ListConfigEntity.PowerUseAllowed` are both set, `InteractiveLaunchSpecService.
+ComposeHandlerPromptAsync` appends `PromptKind.PowerUseAddendum` to whichever handler prompt the
+phase resolved to. The addendum is written as an *override* of the base prompt's interactive rules
+(never ask, decide and record instead; resolve or abort conflicts rather than leaving them in the
+tree; harvest new tasks when a round would starve). Both keys are required on purpose: the global
+switch can stay on overnight without turning an unrelated list's handler run unattended. A run that
+is not in PowerUse gets a byte-identical prompt to the pre-PowerUse behaviour — there is deliberately
+no second set of handler prompts to keep in sync.
+
 Each handoff hands the SAME handler task id to a NEW tile, and `MissionControlViewModel.
 OpenMergeHelperHandoffConPtySessionAsync` closes the outgoing phase's pane (`CloseConPtySession`,
 which disposes its `PtyTerminalSession` and kills the underlying `claude` process) before opening