service-omni-permission-set-assign
Assign one or more PermissionSets to N agent users via PermissionSetAssignment Data API POSTs. The default target is Omni_Agent, the permission set granting the OmniChannel widget, presence-status access, and demo-queue visibility — without an assignment row, agents cannot open the widget or receive routed work. Detection is SOQL-based, so the skill only creates the assignments that are missing. Agent users come from service-omni-agent-users-create, and it runs alongside service-omni-presence-status-deploy so assigned agents can both open the widget and select a status.
Inputs
bash scripts/verify-and-assign.sh <org-alias> [count=3] [permission-set-names-csv=Omni_Agent]
org-alias (required).
count (optional, default 3, range 1..10) — must match the agent user count.
permission-set-names-csv (optional, default Omni_Agent) — comma-separated for multiple. Every user gets every listed set (cross-product): count=3 × 2 sets = up to 6 assignments.
Preconditions and safety
- Target org authenticated via
sf CLI, Service Cloud license, sf CLI ≥ 2.139.6.
- The agent users exist (
service-omni-agent-users-create); a count mismatch blocks with a pointer back to that skill.
- The executing user has
PermissionsAssignPermissionSets (standard on System Administrator) — required even for org admins; ModifyAllData alone is insufficient.
- At least one Omni presence status exists before
Omni_Agent self-heals — the bundled set grants agent capability via servicePresenceStatusAccesses for whichever curated statuses exist (e.g. Available_Case/Available_Voice + Busy), generated at deploy time. Run service-omni-presence-status-deploy first; the coordinator sequences presence before permset for this reason.
- The three-way
safe_to_write guard applies — assigning permission sets on a production org can escalate a real user's privileges, so it blocks with no override.
Self-heal (run mode only). When the default Omni_Agent set is absent, the skill deploys the bundled Omni_Agent metadata once, then assigns. In --plan mode it never deploys — it reports action_needed and exits read-only. This covers only the bundled Omni_Agent asset; any other permission set must already exist or the run blocks with a click-path.
Run
verify-and-assign.sh performs the whole cycle:
- Compute
safe_to_write; derive the 8-char org suffix.
- Resolve the agent users by the
agent{1..N}.<suffix>@example.com pattern; block if any are missing.
- Resolve each
PermissionSet by name; self-heal Omni_Agent if absent, else block naming which is missing.
- Query existing
PermissionSetAssignment for the (user × set) cross-product; compute the missing pairs.
- POST one assignment per missing pair (individual POSTs, no
allOrNone).
- Re-query to confirm final state and emit the report.
Behavior
Cross-product. Every user gets every listed set; a partial assignment is a failure, not a feature.
Idempotency. PermissionSetAssignment has a database uniqueness constraint on (AssigneeId, PermissionSetId), so a re-POST raises DUPLICATE_VALUE; the skill detects existing pairs first and treats DUPLICATE_VALUE as a safety net for concurrent races. It POSTs individually so one duplicate or error never rolls back its successful siblings, and it re-queries after all POSTs — a 201 only means Salesforce accepted the write; a subsequent SOQL confirms the assignment is active.
Non-destructive. The skill is create-only; it never deletes existing assignments (users may hold out-of-band permissions from other admins) and derives users from the agent pattern rather than accepting an explicit user-id list, so it never assigns demo permissions to real named users.
Output contract
A single JSON object with status ∈ assigned | reused | partial | blocked, the resolved permission_sets, org_suffix, requested_count, expected_assignment_count (= requested_count × len(permission_sets)), a before snapshot, assigned_this_run/assigned_count, reused_count, an after snapshot, manual_actions, and blocking_issue.
assigned — at least one new assignment created; all expected pairs exist after.
reused — all expected pairs already existed; nothing POSTed.
partial — some POSTs failed; final count is below expected.
blocked — precondition failed (production org, missing set, missing users, missing permissions).
assigned_count + reused_count == expected_assignment_count unless partial; blocking_issue is non-null only for blocked/partial.
Limitations
- Assigns individual PermissionSets only — a
PermissionSetGroup is a different sObject and is out of scope.
- Self-heal covers only the bundled
Omni_Agent; it is not a general-purpose permission-set authoring surface.
- Create-only; it does not remove assignments.
References