For the complete documentation index, see llms.txt. This page is also available as Markdown.
Command Policy
Command Policies let administrators define, once, which shell commands an agent is allowed to auto-run and which are always denied — enforced platform-wide, per workspace, or per project, rather than set up manually on every ticket. It's a Definition (the regex rules) paired with a Mapping (where those rules apply).
Go to AI Admin → Access Control → Command Policy.
Command Policy tab
Command Policy Definition
A Definition holds two lists of regular expressions — one per line:
Allow List — commands matching any pattern here are auto-approved
Block List — commands matching any pattern here are always denied, even if they also match an allow pattern
Click + Add and fill in:
Name* — a label for the policy
Description — optional
Allow List — one regex per line
Block List — one regex per line
Add Command Policy form filled in
Command Policy added
The Block List always wins — if a command matches both an allow and a block pattern, it's denied. A command that matches neither list isn't auto-decided either way; it falls through to manual approval in the ticket, the same Command Permissions approval prompt described in Command Permissions.
Command Policy Mapping
A Mapping binds a Definition to the scope it should apply to.
Command Policy Mapping tab
Click + Add and fill in:
Name* — a label for the mapping
Description — optional
Command Policy* — the Definition to apply
Level* — System, Workspace, or Project
Target Workspace(s)* / Target Project(s)* — shown only for Workspace/Project level; select one or more
Add Command Policy Mapping form — selecting a policy and level
Add Command Policy Mapping form — target workspace selected
Command Policy Mapping added
Only one active mapping is allowed per level/target combination — for example, only one active System-level mapping platform-wide, and only one active mapping per workspace at the Workspace level.
Command Policies are an admin-configured layer on top of the per-ticket Command Permissions patterns — both are combined when a command is evaluated. A command is denied if it matches a block pattern anywhere (the ticket's own patterns, or the System/Workspace/Project policy covering it); it's auto-approved only if it matches an allow pattern everywhere it's checked. Anything else is left for a person to approve on the ticket.
