diff --git a/AGENTS.md b/AGENTS.md
index 91483965..4bd3a7da 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -85,6 +85,12 @@ every edit is the main way an agent turns a 5-minute task into a 30-minute one,
 
 This section is the *mechanics*. **What to actually run is [`technical-documentation/testing/manual-e2e-checklist.md`](technical-documentation/testing/manual-e2e-checklist.md)** — the capture-to-export pass, per-platform sections, and a results log to append to. Run it before promoting a release candidate and after any change to native capture, preview or export. For cursor work specifically, [`native-cursor-diagnostics.md`](technical-documentation/testing/native-cursor-diagnostics.md) gets you sidecars and reports without a full record-edit-export cycle. The checklist links back here for the mechanics below; the pairing only works if you know both halves exist.
 
+**If the computer-use MCP is available to you, this pass is yours to run — offer it, don't hand it back.** "Manual" names the input (real OS mouse and keyboard events), not the operator; an agent holding that mouse satisfies every check in the file. So when the work calls for it — a change to native capture, preview or export, or a release candidate waiting on QA — say you can run it and run it, instead of closing with "manual e2e pass required" and leaving a maintainer to do what you were already equipped to do. `list_granted_applications` answers whether the MCP is there at all; if it is not, say *that* plainly, because then the checklist genuinely does fall to a human.
+
+**Run whatever slice was asked for.** The whole file is one option, not the only one: a single section, one platform block, or the three checks that cover the change you just made are all legitimate runs — every row in the results log so far is a partial. What is not legitimate is silence about the rest. A check you did not run is **skipped with its reason**, never passed, and the run only exists once it is a row in the results log carrying the build/tag, the platform, and what you did not cover.
+
+**A release candidate is the exception.** The checklist asks for the whole file before a promote, in its own opening, and that is not softened by anything above: a slice covers a *change*, never a promotion. Scope one down and the row you write says `Partial`, which is not a green light to dispatch `promote.yml`.
+
 **Launch the app**
 
 - Normal: `npm run dev` — Vite serves the renderer and `vite-plugin-electron` opens the Electron window. The main process logs `Global shortcut registered: CommandOrControl+Shift+O` when ready (Ctrl/Cmd+Shift+O toggles the HUD).
diff --git a/technical-documentation/engineering/release-and-secrets.md b/technical-documentation/engineering/release-and-secrets.md
index 4cde4187..0e794959 100644
--- a/technical-documentation/engineering/release-and-secrets.md
+++ b/technical-documentation/engineering/release-and-secrets.md
@@ -32,6 +32,8 @@ Run `Promote RC to stable release` (`promote.yml`) with:
 
 The workflow validates the tag, closes the version milestone, checks out `release/vX.Y.Z`, changes `package.json` to the stable version, tags that branch tip, opens and rebase-merges a release-sync PR into `main`, explicitly dispatches `build.yml` at the stable tag, and announces the stable release. The build publishes signed/notarized artifacts when Apple credentials are complete; publication with `OPENSCREEN_RELEASE_TOKEN` emits the event that starts stable Homebrew, WinGet, Nix, and AUR workflows.
 
+**Before dispatching it, the RC has to have been through [the manual end-to-end checklist](../testing/manual-e2e-checklist.md).** Nothing in `promote.yml` enforces that — it will promote an untested tag exactly as readily — so the gate is the operator. It is not a human-only gate either: the checks need real OS mouse and keyboard events, which an agent with the computer-use MCP supplies, so asking one to run it is a normal way to get it done. For a promote that means the **whole file**, not a risk-picked subset — the checklist asks for it in its own opening, and its results log shows why: webcam PiP, microphone, GIF and the AI sections have been scoped out of every run recorded so far, so "the sections this RC puts at risk" is precisely the judgement that keeps missing them. A `Partial` row is not a green light to dispatch. Mechanics and the rule on partial runs: [AGENTS.md](../../AGENTS.md#desktop-e2e-testing-with-computer-use).
+
 ### Release branches (the contract)
 
 Every released version has **exactly one frozen branch**, named for the stable version, living from the first RC cut onward:
diff --git a/technical-documentation/testing/manual-e2e-checklist.md b/technical-documentation/testing/manual-e2e-checklist.md
index 4be4b4b0..dd5c2630 100644
--- a/technical-documentation/testing/manual-e2e-checklist.md
+++ b/technical-documentation/testing/manual-e2e-checklist.md
@@ -2,6 +2,8 @@
 
 This checklist covers the real desktop capture-to-export path: the parts that unit, browser, and Playwright tests cannot exercise, including real screen capture, a physical webcam, the system tray, the native compositor, and export. Run it before promoting a release candidate and after any change to native capture, preview, or export.
 
+**"Manual" is about the input, not the operator.** These checks need real OS mouse and keyboard events, not a human hand — so an agent with the computer-use MCP runs them, on demand, and a request for one section after a targeted change is as much a run as the whole file before a promote. Availability check and the rule on partial runs: [AGENTS.md](../../AGENTS.md#desktop-e2e-testing-with-computer-use).
+
 Sections marked **v1.8.0** cover what this release changed: chat-driven editing through the agent tool set, clip-anchored modifiers, local transcription, the macOS Metal compositor, and the new effect controls. Run the whole file for a release candidate; the v1.8.0 sections are the ones with no prior release to fall back on.
 
 ## How to run this