Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.
Python agentagent-osaiai-agentsai-partnerchatopsclaudeclaude-codecodexdevtoolsdiscord-botlark-botllmlocal-firstlong-term-memoryopencodeslack-botvibe-codingvibeywechat
docs(computer-use): verify activation before retrying no-effect input (#3074)
Max5f296e8f64e3
diff --git a/skills/computer-use/SKILL.md b/skills/computer-use/SKILL.mdindex 07985ee66..94d17c2d6 100644--- a/skills/computer-use/SKILL.md+++ b/skills/computer-use/SKILL.md@@ -56,6 +56,26 @@ its frame, with neither usable visual content nor useful controls, is not readyyet: observe again while the state keeps changing, and report what you observedonce it stops changing or the timeout ends.+## No-effect recovery and viewport content++- Default to background input. If a second observation shows no effect, do not+ repeat the same foreground click immediately: first `await win.activate()`,+ then re-observe that this exact window is on screen and focused before+ retrying that one action. Use `delivery_mode: "foreground"` for the retry if+ background input still has no effect. Foreground input restores the previous+ frontmost window; `activate()` leaves the target frontmost.+- If no on-screen window matches, or activation is unverified, reports+ `focused=false`, or leaves the window off-screen: stop and ask the owner to+ show the window. Do not loop `launch()` or `activate()` or invent a reopen+ path for a hidden window.+- Treat driver `status: unverifiable` as unknown, not proof of effect. After+ input, compare a second observation against a stable signature (title,+ selected row, or visible text). Allow only one retry after verified activation;+ if the effect is still absent or unknown, stop.+- For virtualized content, the tree and screenshot describe the current viewport,+ not the full document. Scrolling replaces that content: observe and capture+ what you need as you scroll; do not assume earlier rows remain in the tree.+## SDK cheat-sheet```js@@ -76,12 +96,9 @@ await avibe.notify("Draft is ready");await avibe.confirm("Send this message to Ana?"); // throws ConfirmRefused on no```-- Input uses background delivery by default and leaves the owner's frontmost- window and pointer alone. Some apps ignore background input: when an- observation shows no effect, repeat that one action with- `delivery_mode: "foreground"`, which fronts the window, acts, and restores the- previous frontmost window. Use `activate()` when the window must stay in- front across several actions; it does not restore the previous window.+- Background delivery leaves the owner's frontmost window and pointer alone.+ For no-effect input, follow the activation and observation procedure above+ before retrying; foreground delivery alone is not proof of effect.- Window selectors: a title shared by off-screen siblings resolves with`onScreen: true`; `windowId` picks an exact id from `observe windows`;`index` orders matches front to back.diff --git a/tests/test_computer_use_backends.py b/tests/test_computer_use_backends.pyindex 6fdd49660..27cc0222e 100644--- a/tests/test_computer_use_backends.py+++ b/tests/test_computer_use_backends.py@@ -105,3 +105,47 @@ def test_computer_use_skill_states_readiness_and_restrictions_not_permissions()for retired in ("require_bind", "Never take focus", "focus_shortcut_forbidden"):assert retired not in bodyassert "verify, `await avibe.confirm(...)`, verify\n again, then act" in body+++def test_computer_use_skill_bounds_no_effect_recovery_and_viewport_evidence() -> None:+ """Guard no-effect recovery; readiness coverage alone misses blind retries."""++ body = Path("skills/computer-use/SKILL.md").read_text(encoding="utf-8")+ first_screen = " ".join("\n".join(body.splitlines()[:80]).split())+ assert "## No-effect recovery and viewport content" in first_screen+ recovery = " ".join(+ body.split("## No-effect recovery and viewport content", 1)[1]+ .split("\n## ", 1)[0]+ .split()+ )+ assert recovery in first_screen+ assert "Default to background input" in recovery++ # A no-effect observation gates activation, exact-window verification, then retry.+ steps = (+ "second observation shows no effect",+ "do not repeat the same foreground click immediately",+ "await win.activate()",+ "re-observe that this exact window is on screen and focused",+ "retrying that one action",+ )+ positions = [recovery.index(step) for step in steps]+ assert positions == sorted(positions)+ assert 'Use `delivery_mode: "foreground"` for the retry if background' in recovery+ assert "Foreground input restores the previous frontmost window" in recovery+ assert "`activate()` leaves the target frontmost" in recovery++ # Missing visibility or unverified focus requires owner recovery, not a loop.+ stop_rule = recovery.split("- If ", 1)[1].split("- Treat ", 1)[0]+ for condition in ("no on-screen window", "unverified", "focused=false", "off-screen"):+ assert condition in stop_rule+ assert "stop and ask the owner to show the window" in stop_rule+ assert "Do not loop `launch()` or `activate()`" in stop_rule++ assert "`status: unverifiable` as unknown, not proof of effect" in recovery+ assert "compare a second observation against a stable signature" in recovery+ assert "only one retry after verified activation" in recovery+ assert "if the effect is still absent or unknown, stop" in recovery+ assert "current viewport, not the full document" in recovery+ assert "observe and capture what you need as you scroll" in recovery+ assert "do not assume earlier rows remain in the tree" in recovery