SuperAPI AI Directory

permission-max

Agents & models

View repository

Updated 10-15 16:46

About this skill

permission-max

Objective: In order to reduce the permission confirmations that occur frequently during a Microsoft Scout session, request the Scout-side shell allow list / read-only auto approve / server autoApprove / tool allow map as broadly as possible, and also diagnose cases where confirmations such as Read / Write / PowerShell command on the Copilot CLI / host side appear.

When to Use

  • "permission max", "want to reduce permission confirmations", "want to expand the shell allow list"
  • Scout の server/tool autoApprove want to check/expand
  • Read / Write / Edit / PowerShell command confirmations are repeated, and want to determine whether Scout or the CLI / host is controlling them
  • For another process's copilot -p , want to check the available allow flags

Important assumptions

  • This Skill does not bypass security confirmations. Always check the current settings, and request only what can be changed, with user approval.
  • Treat Scout-side permissions and Copilot CLI / host-side tool execution confirmations as separate layers.
  • Allow everywhere It does not guarantee complete unconditional permission equivalent to, or that security-sensitive location warnings will be removed. If warnings remain, treat them as safety guards.
  • For path permissions, within the scope Scout permits, prioritize confirmed parent folders/wildcards such as <allowed-root>\* . However, m_request_permission_escalation mainly changes the shell allow list, tool allow map, autoApproveReadOnly, and server autoApprove, and it is not always possible to change path-level warnings all at once.
  • allow is a full replacement, so always read the current allow list and pass the union of existing items + additional items as string[] . Do not drop existing items.
  • If no setting is found that reduces CLI / host-side confirmations, explicitly state that "the Scout side has been maximized, but the CLI side is a separate guard and cannot be changed." Do not say "done" based on guesswork.
  • For dangerous operations, deletions, external transmissions, M365 sends, sign-outs, service stops, etc., prioritize the user's explicit instructions and confirmation even if permissions are broad.

Target layers

Layer Example What to do with this Skill
Scout permissions m_get_settings の permissions.allow / tools / servers m_request_permission_escalation request approval with
Scout file/server autoApprove filesystem / shell / playwright / workiq request to enable server autoApprove
Copilot CLI / host tool confirmation Read / Write / Edit / PowerShell command confirmation of Investigate the setting location and configure it if changeable. If not possible, report as a limitation
command allow flags copilot -p ... --allow-all-tools etc. Subprocess's copilot Record as a workaround usable only when running

Procedure A: Maximizing the Scout side

  1. m_get_settings and check the following.

    • permissions.allowModelPermissionsChange
    • Current permissions.allow
    • Current permissions.tools
    • Current permissions.servers
    • permissions.autoApproveReadOnly
  2. permissions.allowModelPermissionsChange が false If, stop and tell the user to enable Settings -> Permissions -> "Allow AI to request permission changes" and then run again.

  3. Scout permission profile and create a deduplicated union of the current allow list and the requested patterns.allow is a full replacement, so do not drop existing items.

  4. Create an escalation payload containing only server/tool IDs that exist in the current environment, and with user approval call m_request_permission_escalation .

  5. After requesting, run m_get_settings again and, ignoring array order, added / preserved / removed / not-approved Classify as.removed If there is even one, do not treat it as successful.

Execution procedure B: Diagnose Copilot CLI / host-side confirmations

Even if the Scout side has been maximized,Read / Write / Edit / PowerShell command if a confirmation appears, check the following.

  1. First, explain to the user that "this may be a separate layer on the Copilot CLI / host side rather than the Scout side."

  2. Copilot CLI and host diagnostics In accordance with, investigate candidate settings and the current CLI help in a read-only manner.

  3. Only if a corresponding approval / confirmation / workspace trust / tool allow-list setting is found, explain the scope to the user before editing.

  4. If the setting is not found, or cannot be changed due to host management, report that limitation and the scope of application of subprocess flags separately.

Execution procedure C: Assistance when running the subprocess copilot

It may not be effective for confirmations in the current conversation. The available flags and their scope of application should be checked against the current CLI help with Copilot CLI and host diagnostics .

Reporting policy

  • Always report "Scout side" and "Copilot CLI / host side" separately.
  • Clarify which one was maximized and which one could not be changed.
  • Do not say "maximization complete" when no settings have been changed.
  • If only a read-only investigation was completed, report it as "investigation results."
  • Scout results are indicated by added / preserved / removed / not-approved , andremoved if there is, do not treat it as complete until it is repaired.