extension.js · part 20

60 text occurrences from desktop/Cursor.app/Contents/Resources/app/extensions/cursor-computer-use/dist/extension.js, part 20. Every entry preserves the shipped literal and its saved verdict or selection reason.

File contents and all parts · All files

${t.productName} manifest artifact and executable disagree

${t.productName} artifact URL is not content-addressed

${io(n.failure)}${n.failure.retryable?"; retried on the next tool call":""}

; retried on the next tool call

Cursor ${this.options.hostVersion} predates the ${ri.productName} host floor ${Dr}; update Cursor

${ri.productName} publishes x64 and arm64 builds only; this Cursor is ${t}

another Cursor window is installing ${ri.productName}; try again in a moment

the ${ri.productName} release config could not be read; using the verified cached release ${hi(n.manifest)} at ${n.resolved.executablePath}

the ${ri.productName} release config read empty for ${a} (the Statsig fallback, or no release is configured); no download is attempted and the ${n.source} sidecar at ${n.executablePath} is kept

no ${ri.productName} sidecar is installed and the release config could not be read

no ${ri.productName} sidecar is installed and the release config read empty for ${a} (the Statsig fallback, or no release is configured for this build)

Ignoring the ${ri.productName} CDN manifest in Cursor Dev; using the ${n.source} sidecar at ${n.executablePath}

the downloaded ${ri.productName} sidecar requires a signed Labs or production Cursor client; Cursor Dev needs a local sidecar in ${p.join("%USERPROFILE%",...xs)} or CUA_SERVICE_EXE

the configured ${ri.productName} release is not an exact immutable Windows release for ${a}

another Cursor window is installing ${ri.productName}; try again in a moment

${io(i.failure)}; falling back to the ${n.source} sidecar at ${n.executablePath}

the configured ${ri.productName} release ${hi(i.manifest)} is already installed for ${t}; nothing to download

cached ${ri.productName} sidecar is kept but not launched: ${lo(e)}

cached ${ri.productName} sidecar failed verification and will be replaced: ${lo(e)}

the downloaded ${ri.productName} sidecar did not match its published SHA-256

the ${ri.productName} sidecar could not be written to ${p.dirname(e)}

the ${ri.productName} sidecar could not be moved into ${e}

${JSON.stringify(c)}

${JSON.stringify(i)}

installed ${ri.productName} sidecar ${hi(c)} for ${t}

the ${ri.productName} release config could not be read; a verified cached release or the development sidecar is used instead: ${lo(e)}

${ri.productName} release for ${e}, from the Statsig config: ${t}

cached ${ri.productName} sidecar is not usable while the release config is unreadable: ${lo(e)}

the ${ri.productName} install lock could not be released; other windows wait for this one to exit or for the lock to go stale: ${lo(e)}

using the cached ${ri.productName} sidecar ${hi(n.manifest)}; release refresh failed: ${io(o)}

cached ${ri.productName} sidecar failed verification: ${lo(e)}

not a regular file

the ${ri.productName} sidecar at ${e} could not be read

the ${ri.productName} sidecar at ${e} does not match its manifest

no MZ header

no PE signature

machine 0x${a.toString(16)} is not ${n} (0x${s.toString(16)})

not a PE32+ (64-bit) image

the ${ri.productName} sidecar at ${e} is not a Windows ${t} executable

the ${ri.productName} sidecar at ${e} could not be checked for its Authenticode signature; it is not launched until the check succeeds

the ${ri.productName} sidecar at ${e} is ${function(e){switch(e.kind){case"trusted":case"untrusted-signer":return`signed by ${e.subject}`;case"unsigned":return`not validly signed (${e.detail})`;case"unverifiable":return`unverifiable (${e.detail})`}}(s)}; only a build signed by ${ri.pinnedSignerSubjects.join(" or ")} is launched

signed by ${e.subject}

not validly signed (${e.detail})

unverifiable (${e.detail})

${e.sidecarVersion} (${e.lane}, ${e.sourceSha.slice(0,12)})

the published ${ri.productName} release manifest was refused

installing the ${ri.productName} sidecar failed

the ${ri.productName} install record could not be written to ${p.dirname(e)}

- Call computer_check_permissions first and wait for its result before any other Computer Use tool; do not call Computer Use tools in parallel, they act on one machine and run one at a time. Continue only when both permissions are granted.

- If a tool reports a missing permission (macOS asks for Accessibility and Screen Recording), run computer_check_permissions and ask the user to grant it in System Settings.

These tools drive one macOS app in the background. They do not take the display or move the real cursor.

- Default to an app target: call computer_apps or computer_resolve_app, then pass {"scope":"app","pid":…,"target_id":…} on every action. Do not call computer_start_control first.

- Observe with computer_screenshot (app target) for the image. Call computer_app_state when you need element ids (first observe, or a screenshot that omitted snapshot_id). Skip computer_app_state when a screenshot reused a live snapshot_id. computer_wait only delays; it returns no image. Prefer computer_set_value or computer_type with element_id. Without element_id, computer_type inserts into the click-selected field via AX, or the focused field. A missed click target is refused. If nothing is focused, click a field first. Does not raise or move the real cursor. Screen-scope typing is refused.

- Spotlight, Notification Center, and menu extras are not listed by computer_apps and are unsupported.

- computer_start_control is escalation only: it takes the Mac and moves the real cursor. Use it only when the user asks to take over the display. After screen-scope work, call computer_release_control.

These tools control the local desktop: they read the real screen and send real mouse and keyboard input.

- Look before you click: take a screenshot before your first coordinate action, and rely on the fresh screenshot every action returns instead of assuming the screen is unchanged.

- Before the first input action, call computer_start_control. A later explicit start is required after the user stops control or the helper restarts.

- Every input action returns a screenshot of the screen after the action. If the result is not what you expected, take another screenshot before retrying.

- After the final input action, call computer_release_control as cleanup.