Update to Desktop Alsalibi
Snapshot Nov 3, 2026 · 19:23 UTC · version 1.2.7
Collection source: downloaded plugin package.
Instructions updated for desktop-alsalibi
Full description — Now says: a user's paired Windows, macOS or Linux computers for device health, approved commands, files, processes, screens and
Observed in instructions or declared skills. Runtime behavior has not been tested.
Up to 10 plugins free · no card required.
All fields and original values.
Full description
Oct 31, 2026 · 06:10 UTC
the user's paired computers when asked to list devices, inspect health, run terminal commands, read or write files, inspect processes, capture screens, or perform supported
Nov 3, 2026 · 19:23 UTC
a user's paired Windows, macOS or Linux computers for device health, approved commands, files, processes, screens and
Skill instructions
Oct 31, 2026 · 06:10 UTC
the user's paired computers when asked to list devices, inspect health, run terminal commands, read or write files, inspect processes, capture screens, or perform supported desktop actions. Use the configured Desktop Alsalibi MCP server ...
Nov 3, 2026 · 19:23 UTC
a user's paired Windows, macOS or Linux computers for device health, approved commands, files, processes, screens and desktop actions. 1. Start with `list_devices` and use the exact device ID returned for the authenticated account. Never...
Compare saved observations
Download comparison JSONSKILL.md text changes
− removed · + added
--- before +++ after @@ -1,6 +1,15 @@ --- name: desktop-alsalibi -description: Use Alsalibi Connect to work with the user's paired computers when asked to list devices, inspect health, run terminal commands, read or write files, inspect processes, capture screens, or perform supported desktop actions. +description: Use Alsalibi Connect to work with a user's paired Windows, macOS or Linux computers for device health, approved commands, files, processes, screens and desktop actions. --- -Use the configured Desktop Alsalibi MCP server for computer-control requests. Start with `list_devices` and use the exact device ID returned by the server. If a tool returns an OAuth authentication challenge, let ChatGPT connect or reconnect the user's Desktop Alsalibi account, then retry the tool. Prefer read-only checks before changes unless the user explicitly requested the change, and respect device-level permissions. +1. Start with `list_devices` and use the exact device ID returned for the authenticated account. Never infer a device ID from a hostname, and do not treat a stale device record as proof of an online computer. +2. Distinguish login state from device state. If a tool returns "Authentication required", 401, or a token/session error, use the host-managed OAuth reconnect flow and retry once. Do not tell the user that all computers are offline based on an authorization error. If the reconnected route still fails, report the route and error without guessing. +3. Use `get_profile` when needed to verify the intended account. Different installed public/private versions can share a display name while having different connections; never silently switch account identities. +4. When the user requests a supported action, use the connected tool rather than sending installation or terminal commands for the user to copy. Check the device and permissions first when warranted; prefer read-only inspection before modifications. +5. Distinguish the ChatGPT plugin package version (such as 1.2.7) from the separate desktop agent version reported for each device. Do not assume an agent upgrade occurred merely because the plugin package changed. +6. The Windows terminal runs commands through cmd.exe by default; use CMD syntax or explicitly call `powershell.exe -NoProfile -Command` for PowerShell operations. macOS/Linux use their actual shell; check command availability and fall back appropriately (for example, `grep` when `rg` is not installed). +7. For reconnect or auto-start faults, inspect the specific machine: Windows background service/task, macOS service or launch agent, or Linux systemd service, as installed. Verify results; do not claim persistent startup from a single successful call. +8. Respect each operating system's and user's approvals. Advanced Control is optional, installed only by the computer owner with administrator authorization. Never bypass UAC, macOS TCC/SIP, Linux privilege policy, ChatGPT permission prompts or screen locks. +9. Keep GUI and keyboard operations in the active desktop session. Use privileged helpers only for tasks explicitly authorized by the owner and supported by the installed agent. +10. Ask before destructive actions, major security-setting changes, or public deployment. Never assume a paired device grants blanket authority to change unrelated files or accounts. Report partial outcomes and errors accurately.
Full technical diff · 2 changed fields
changed /description
"Use Alsalibi Connect to work with the user's paired computers when asked to list devices, inspect health, run terminal commands, read or write files, inspect processes, capture screens, or perform supported desktop actions."
"Use Alsalibi Connect to work with a user's paired Windows, macOS or Linux computers for device health, approved commands, files, processes, screens and desktop actions."
changed /skill_md_contents
"---\nname: desktop-alsalibi\ndescription: Use Alsalibi Connect to work with the user's paired computers when asked to list devices, inspect health, run terminal commands, read or write files, inspect processes, capture screens, or perform supported desktop actions.\n---\n\nUse the configured Desktop Alsalibi MCP server for computer-control requests. Start with `list_devices` and use the exact device ID returned by the server. If a tool returns an OAuth authentication challenge, let ChatGPT connect or reconnect the user's Desktop Alsalibi account, then retry the tool. Prefer read-only checks before changes unless the user explicitly requested the change, and respect device-level permissions.\n"
"---\nname: desktop-alsalibi\ndescription: Use Alsalibi Connect to work with a user's paired Windows, macOS or Linux computers for device health, approved commands, files, processes, screens and desktop actions.\n---\n\n1. Start with `list_devices` and use the exact device ID returned for the authenticated account. Never infer a device ID from a hostname, and do not treat a stale device record as proof of an online computer.\n2. Distinguish login state from device state. If a tool returns \"Authentication required\", 401, or a token/session error, use the host-managed OAuth reconnect flow and retry once. Do not tell the user that all computers are offline based on an authorization error. If the reconnected route still fails, report the route and error without guessing.\n3. Use `get_profile` when needed to verify the intended account. Different installed public/private versions can share a display name while having different connections; never silently switch account identities.\n4. When the user requests a supported action, use the connected tool rather than sending installation or terminal commands for the user to copy. Check the device and permissions first when warranted; prefer read-only inspection before modifications.\n5. Distinguish the ChatGPT plugin package version (such as 1.2.7) from the separate desktop agent version reported for each device. Do not assume an agent upgrade occurred merely because the plugin package changed.\n6. The Windows terminal runs commands through cmd.exe by default; use CMD syntax or explicitly call `powershell.exe -NoProfile -Command` for PowerShell operations. macOS/Linux use their actual shell; check command availability and fall back appropriately (for example, `grep` when `rg` is not installed).\n7. For reconnect or auto-start faults, inspect the specific machine: Windows background service/task, macOS service or launch agent, or Linux systemd service, as installed. Verify results; do not claim persistent startup from a single successful call.\n8. Respect each operating system's and user's approvals. Advanced Control is optional, installed only by the computer owner with administrator authorization. Never bypass UAC, macOS TCC/SIP, Linux privilege policy, ChatGPT permission prompts or screen locks.\n9. Keep GUI and keyboard operations in the active desktop session. Use privileged helpers only for tasks explicitly authorized by the owner and supported by the installed agent.\n10. Ask before destructive actions, major security-setting changes, or public deployment. Never assume a paired device grants blanket authority to change unrelated files or accounts. Report partial outcomes and errors accurately.\n"
Full snapshot data
{
"description": "Use Alsalibi Connect to work with a user's paired Windows, macOS or Linux computers for device health, approved commands, files, processes, screens and desktop actions.",
"included_files": [],
"name": "desktop-alsalibi",
"skill_md_contents": "---\nname: desktop-alsalibi\ndescription: Use Alsalibi Connect to work with a user's paired Windows, macOS or Linux computers for device health, approved commands, files, processes, screens and desktop actions.\n---\n\n1. Start with `list_devices` and use the exact device ID returned for the authenticated account. Never infer a device ID from a hostname, and do not treat a stale device record as proof of an online computer.\n2. Distinguish login state from device state. If a tool returns \"Authentication required\", 401, or a token/session error, use the host-managed OAuth reconnect flow and retry once. Do not tell the user that all computers are offline based on an authorization error. If the reconnected route still fails, report the route and error without guessing.\n3. Use `get_profile` when needed to verify the intended account. Different installed public/private versions can share a display name while having different connections; never silently switch account identities.\n4. When the user requests a supported action, use the connected tool rather than sending installation or terminal commands for the user to copy. Check the device and permissions first when warranted; prefer read-only inspection before modifications.\n5. Distinguish the ChatGPT plugin package version (such as 1.2.7) from the separate desktop agent version reported for each device. Do not assume an agent upgrade occurred merely because the plugin package changed.\n6. The Windows terminal runs commands through cmd.exe by default; use CMD syntax or explicitly call `powershell.exe -NoProfile -Command` for PowerShell operations. macOS/Linux use their actual shell; check command availability and fall back appropriately (for example, `grep` when `rg` is not installed).\n7. For reconnect or auto-start faults, inspect the specific machine: Windows background service/task, macOS service or launch agent, or Linux systemd service, as installed. Verify results; do not claim persistent startup from a single successful call.\n8. Respect each operating system's and user's approvals. Advanced Control is optional, installed only by the computer owner with administrator authorization. Never bypass UAC, macOS TCC/SIP, Linux privilege policy, ChatGPT permission prompts or screen locks.\n9. Keep GUI and keyboard operations in the active desktop session. Use privileged helpers only for tasks explicitly authorized by the owner and supported by the installed agent.\n10. Ask before destructive actions, major security-setting changes, or public deployment. Never assume a paired device grants blanket authority to change unrelated files or accounts. Report partial outcomes and errors accurately.\n"
}SHA-256 of public snapshot: b3a50fd5785142b6db175b3d14dd69e97fa88e2c82102c23ff4e30044b8e8622