Skills

Manage reusable instruction bundles that extend agent capabilities without modifying each agent's core instructions.

Overview

Skills are named instruction sets stored in Cosmos DB. When an agent executes, the skill manifest service merges the tenant's active skills into the agent's prompt, giving every agent access to shared capabilities like email formatting patterns, escalation logic, or compliance workflows.

Plan Availability

Company-wide skills (Tenant scope - shared across every agent) require the Workforce plan or higher. On Outermind Assistant, creating, editing, or deleting a company-wide skill is refused; built-in skills and a Personal Assistant's own Agent-scope skills remain available on every plan.

If a tenant downgrades from Workforce to Assistant, its existing company-wide skills go dormant: they stop being injected into agent prompts, but the records are preserved and resume automatically on the next upgrade. Turning a built-in skill off and on (the global override toggle) is not affected by this gate.

Skills have a lifecycle status that controls whether they appear in the LLM's prompt:

Status	Prompt Visible	Editable
Active	Yes	Yes
Experimental	No	Yes
Draft	No	Yes
Pending Approval	No	Read-only
Deprecated	No	Yes (can reactivate)
Disabled	No	Yes

Navigating to Skills

Open the Capabilities Hub from the top navigation, then select Skills in the sidebar.

Skills List

The Skills List shows all tenant skills in a table with the following columns:

• Name - Unique kebab-case identifier (e.g. email-tone-guide)
• Scope - Tenant (all agents) or Agent (specific agent)
• Status - Color-coded badge (see table above)
• Activation Mode - How the skill is injected: Auto, On-Demand, or Hybrid
• Updated - Last modification date
• Proposed By - Agent that proposed the skill (if auto-generated)

Filtering

Use the Scope and Status dropdowns in the toolbar to narrow the list.

Actions

• Click any row to open the Skill Editor for that skill.
• Use New Skill to create a skill manually.
• Use View Proposed (with badge count) to review agent-proposed skills awaiting approval.

Deprecating a Skill

Deprecation removes a skill from the LLM's prompt without deleting the record, preserving audit history.

1. From the Skills List, locate the skill you want to deprecate.
2. Click Mark Deprecated in the row actions. A confirmation dialog appears.
3. Confirm the action.

The skill's status changes to Deprecated (amber badge) and it is immediately excluded from prompt injection and from tool_search results.

A deprecated row shows a Reactivate button in place of the deprecation action.

Reactivating a Deprecated Skill

1. Locate the deprecated skill (filter by Status = Deprecated).
2. Click Reactivate. A confirmation dialog appears.
3. Confirm the action.

The skill's status returns to Active and prompt injection resumes on the next agent execution.

Creating and Editing Skills

Click New Skill or an existing skill row to open the Skill Editor.

Basic Info

Field	Description
Name	Kebab-case identifier, unique within the tenant
Display Name	Human-readable label shown in the console
One-liner	Short description (max 80 characters)

Instructions

Enter the markdown-formatted instructions the skill injects into the agent prompt. Keep instructions focused on a single capability to simplify future deprecation.

Tool References

Comma-separated list of tool names the skill's instructions reference. Used for dependency tracking.

Activation

Setting	Description
Mode	Auto (always injected), On-Demand (injected when triggered), Hybrid
Triggers	Patterns that activate the skill (auto/hybrid modes)
Profiles	Context profiles that enable this skill
Context Match	Free-text context matching rule
Context Tags	Tags used to match conversation context

Status in Edit Mode

In the Skill Editor, the Status dropdown allows the following transitions for existing skills:

• Draft, Active, Experimental, Deprecated, Disabled

Pending Approval is shown read-only when the skill is awaiting human review; no other transition is available until the skill is approved or rejected.

For an agent's Inbox Triage skill, changing an active skill's instructions or name, or setting it to Active, requires the supervisor whose mailbox the rules manage; anyone else sees Inbox Triage changes require the current supervisor. Administrators can still disable it.

Proposed Skills

Agent-proposed skills appear on the Proposed Skills page (/capabilities/skills/proposed). Each card shows the skill name, description, proposing agent, and a preview of the instructions.

• Approve - Creates the skill as Active. An agent's Inbox Triage proposal can be approved only by the supervisor whose mailbox its rules manage; anyone else sees Inbox Triage approval requires the current supervisor.
• Reject - Removes the proposal; a rejection reason is required.

Managing Inbox Triage Rules

Inbox triage rules guide how your Chief of Staff or Personal Assistant handles incoming mail. Open Inbox Triage from the Boardroom or PA Hub, then find Your rules. The page shows the active rules and any proposed changes awaiting your approval.

Which Incoming Messages Reach Your Rules?

Incoming mail in both Outlook Focused and Other tabs can reach your approved Inbox Triage rules. A message is no longer skipped simply because Outlook filed it in Other. You do not need to move it to Focused first.

Other filters still apply before rules are evaluated. Bulk mail marked with List-Unsubscribe or Precedence: bulk, list, or junk is skipped, as are calendar messages, automated submissions, no-reply mail, and mail from the platform or its agents. A rule such as "move solicitations to junk" can therefore handle eligible Other-tab mail, but does not override these filters for newly arriving messages.

Forwarding to an Address Named in Your Rules

Your Chief of Staff or Personal Assistant can forward an original message from your mailbox, including its attachments, to an external email address named in your approved active rules. Your approval authorizes that address for Inbox Triage forwarding while it remains in the active rules. You do not need to approve each matching forward separately.

1. Open Inbox Triage from the Boardroom or PA Hub.
2. Select Make changes or Change in chat and describe when to forward a message. Include the exact destination email address, for example: "When a message is confirmed as a supplier invoice, forward it to invoices@example.com."
3. Review the proposed rules, including every external email address they contain. Approval authorizes the external addresses named in the rules.
4. As the supervisor whose mailbox is being managed, select Approve to activate the rules.

The agent needs Mailbox Read & Write under Settings > Data access (PA Hub) or Boardroom > Settings > Data access to apply triage rules. Pending proposals do not authorize forwarding. Recipients absent from the active approved rules are refused, and triage forwards cannot add Cc or Bcc recipients. Every forward still passes through the Safety Gateway and can be blocked by its checks.

This permission applies only to Inbox Triage forwarding. It does not authorize other unattended sends. The resulting Sent Items notification does not start another triage run or create a new activity signal.

To withdraw permission for an address, delete its active rule or approve revised rules that remove the address. Permission remains if another active rule still names that address. Disabling the Inbox Triage skill or removing its last active rule also withdraws its forwarding permissions. Revoking the address on Standing authorizations does not last while a rule still names it: the permission is restored the next time Inbox Triage runs, so change the rules instead.

Daily Limits

Inbox Triage uses a per-agent daily cost budget for screening and runs, defaulting to $5. It resets at midnight in the tenant's timezone.

Administrators can edit Max Daily Inbox Triage Cost ($) in Usage Limits in the Personal Assistant's Spending limits section or the AICOS Spending limits section. The amount must be at least $1 and must not exceed Max Daily Usage Cost ($). Select Save Usage Limits to apply a change; this budget is not edited on the rules page.

Deleting a Rule

1. Find the rule in the Active list or under Proposed change, waiting for you.
2. Select Delete beside that rule.
3. Review the rule text in the Delete this rule? dialog.
4. Select Delete rule to confirm, or Cancel to keep it.

After a successful deletion, the page reloads the rules and renumbers the remaining items. Deletion does not open the chat drawer or require a separate approval.

• Active rule: Your agent stops applying the deleted rule immediately. Any pending proposal stays unchanged, so review it before approving it if it still contains the rule you removed.
• Pending rule: The rule is removed from the proposal. Your active rules keep running; the remaining proposal still needs approval before it takes effect.

Deleting the last active rule returns triage to signals only: incoming mail sends signals to your agent's linked work without applying a triage rule set. Deleting the last pending rule removes that proposal without changing the active rules.

Making Other Changes

Select Make changes beside a rule, or Change in chat above the active list, to open the agent chat drawer. Review the resulting proposal and select Approve when you are ready to activate it. You can also select Reject and provide a reason.

If the Rules Changed While You Were Reviewing Them

If you see Rules changed. Refresh and try again., review the refreshed list before retrying. Rule numbers may have changed; select the intended rule again and check its text in the confirmation dialog.

Frequently Asked Questions

Why was an Inbox Triage forward refused? Check that the exact recipient address appears in the active rules approved by the mailbox's supervisor, rather than only in a pending proposal. Verify Mailbox Read & Write access and ensure the forward has no Cc or Bcc recipients. Approval of a recipient does not bypass Safety Gateway checks.

Will the forward preserve the original attachments? Yes. Inbox Triage forwards the original from your mailbox with its attachments.

Why is my skill not appearing in agent responses? Check the skill's status. Only Active skills are included in the prompt. If the status is correct, verify the skill's Activation Mode and that its triggers match the current context.

Can I recover a deleted skill? Hard deletion is permanent. Use Deprecated status instead of deletion when you want to preserve the skill for audit purposes.

What happens to the prompt cost when I deprecate a skill? Deprecated skills are excluded from buildPromptSection, so their token cost is eliminated immediately on the next agent execution.

How do I find skills that have not been used recently? The platform runs an automated activation-decay check nightly. Skills with no [SKILL_AUTO_ACTIVATED] events in the last 60 days and status: active are flagged in the build pipeline. Review flagged skills and deprecate them if they are no longer needed.

Why can't I create or edit a company-wide skill? Company-wide (Tenant scope) skills require the Workforce plan or higher. Upgrade the subscription, or use an Agent-scope skill on a specific assistant instead. If the tenant was recently downgraded from Workforce, its existing company-wide skills are dormant rather than deleted, and resume once the plan is upgraded again.