Skip to main content

Policy Guards

Policy guards are contextual rules that inspect, filter, or block agent actions in real time. Each guard targets a specific action type (sending email, sharing files, deleting messages, etc.) and applies one of three enforcement actions:

  • deny -- Block the action outright.
  • require_approval -- Pause the action and request human approval before proceeding.
  • filter -- Allow the action but redact or strip sensitive data from the response.

Guards are organized into preset tiers (standard and strict) so operators can adopt a baseline quickly and customize from there.


Messaging

Guards for sending emails, chat messages, channel posts, replies, and forwarding.

Email Sending

Guard IDNameDescriptionRiskProvidersAction
msg-profanity-emailProfanity filter (email)Block outbound emails whose body contains profane or offensive languageHighGoogle, Microsoftdeny
msg-pii-emailPII guard (email body)Block or flag emails containing SSNs, credit cards, or other PII in the bodyHighGoogle, Microsoftrequire_approval
msg-recipient-limit-emailRecipient limit (email)Require approval when sending to more than N recipients (To+Cc+Bcc). Default threshold: 10MediumGoogle, Microsoftrequire_approval
msg-external-domain-emailExternal domain guard (email)Require approval when sending to recipients outside the org domainHighGoogle, Microsoftrequire_approval
msg-attachment-guard-emailAttachment type guard (email)Block emails with executable or script attachments (.exe, .bat, .ps1, .sh, .js)HighGoogle, Microsoftdeny

Chat Sending

Guard IDNameDescriptionRiskProvidersAction
msg-profanity-chatProfanity filter (chat)Block chat messages containing profane or offensive languageMediumMicrosoft, Telegramdeny
msg-pii-chatPII guard (chat)Block or flag chat messages containing PIIMediumMicrosoft, Telegramrequire_approval
msg-link-guard-chatExternal link guard (chat)Flag messages containing URLs to external domainsLowMicrosoft, Telegramrequire_approval

Channel Posts

Guard IDNameDescriptionRiskProvidersAction
msg-profanity-channelProfanity filter (channel post)Block channel/group posts containing profane languageHighMicrosoft, Telegramdeny
msg-pii-channelPII guard (channel post)Block channel posts containing PII. Audience amplification makes PII leaks worse in channelsHighMicrosoft, Telegramdeny
msg-audience-limit-channelAudience size guard (channel)Require approval for posts to channels/groups with many members. Default threshold: 50HighMicrosoft, Telegramrequire_approval

Replies and Forwarding

Guard IDNameDescriptionRiskProvidersAction
msg-profanity-replyProfanity filter (reply)Block reply messages containing profane languageMediumGoogle, Microsoft, Telegramdeny
msg-forward-guardForward message guardRequire approval or block message forwarding. Forwarding can leak context to unintended recipientsMediumTelegramrequire_approval

File Sharing

Guards for sharing, uploading, and sending files and media.

Sharing

Guard IDNameDescriptionRiskProvidersAction
fs-public-share-blockBlock public sharingBlock sharing files with public/anonymous accessHighGoogle, Microsoftdeny
fs-external-share-guardExternal sharing guardRequire approval for sharing with external users/domainsHighGoogle, Microsoftrequire_approval
fs-link-sharing-blockBlock link sharingBlock creating shareable links (domain-wide or public)MediumGoogledeny

Uploading

Guard IDNameDescriptionRiskProvidersAction
fs-file-type-guardFile type guard (upload)Block uploading executable or script files (.exe, .bat, .ps1, .sh, .js, .vbs, .msi, .dll)HighGoogle, Microsoftdeny
fs-file-size-guardFile size guardRequire approval for uploads exceeding size threshold. Default: 25 MBMediumGoogle, Microsoftrequire_approval
fs-sensitive-filenameSensitive filename guardBlock uploads with filenames suggesting credentials or secrets (password*, secret*, credential*, .env, .pem)HighGoogle, Microsoftdeny

Media (Telegram)

Guard IDNameDescriptionRiskProvidersAction
fs-media-type-telegramMedia type guard (Telegram)Restrict which media types can be sent. Approval for document sending; photos/videos may auto-allowMediumTelegramrequire_approval

Calendar

Guards for creating and modifying calendar events.

Guard IDNameDescriptionRiskProvidersAction
cal-external-attendeeExternal attendee guardRequire approval when inviting attendees outside the orgHighGoogle, Microsoftrequire_approval
cal-large-meetingLarge meeting guardRequire approval for events with many attendees. Default threshold: 20MediumGoogle, Microsoftrequire_approval
cal-off-hoursOff-hours booking guardFlag events created outside business hours. Business hours configurable per timezoneLowGoogle, Microsoftrequire_approval
cal-cancel-guardEvent cancellation guardRequire approval before canceling events with 2+ attendeesMediumGoogle, Microsoftrequire_approval

Data Reading

Response-side guards that filter or redact sensitive information from API responses.

Guard IDNameDescriptionRiskProvidersAction
dr-strip-contact-piiStrip contact PIIRemove phone numbers, addresses, and birthdays from contact responsesMediumGoogle, Microsoftfilter
dr-redact-contact-emailRedact contact emailsRedact email addresses in contact/directory responsesMediumGoogle, Microsoftfilter
dr-redact-message-piiRedact PII in messagesRedact PII (SSN, phone, credit card) found in read message responsesMediumGoogle, Microsoft, Telegramfilter
dr-redact-file-metadataRedact file metadata PIIRedact owner PII from file listing responsesLowGoogle, Microsoftfilter
dr-rate-limit-searchSearch rate limitLimit search query frequency to prevent data exfiltrationMediumGoogle, Microsoftallow

Destructive

Guards that protect against data deletion, member removal, and permission revocation.

Message Deletion

Guard IDNameDescriptionRiskProvidersAction
dest-delete-msg-blockBlock message deletionBlock agent from deleting messages. Prevents evidence destruction or disruptionHighMicrosoft, Telegramdeny
dest-delete-msg-approveApprove message deletionRequire approval for message deletion (less strict alternative)MediumMicrosoft, Telegramrequire_approval

File Deletion

Guard IDNameDescriptionRiskProvidersAction
dest-delete-file-approveApprove file deletionRequire approval before deleting filesHighGoogle, Microsoftrequire_approval
dest-delete-file-blockBlock file deletionBlock agent from deleting any files (strictest mode)HighGoogle, Microsoftdeny

Member Removal

Guard IDNameDescriptionRiskProvidersAction
dest-kick-blockBlock member removalBlock agent from kicking/banning membersHighTelegramdeny
dest-kick-approveApprove member removalRequire approval before kicking/banning membersHighTelegramrequire_approval

Permission Revocation

Guard IDNameDescriptionRiskProvidersAction
dest-revoke-perm-approveApprove permission revocationRequire approval before revoking file/resource permissionsMediumGoogle, Microsoftrequire_approval

Administrative

Guards for role management, settings changes, and member additions.

Guard IDNameDescriptionRiskProvidersAction
adm-role-change-blockBlock role changesBlock agent from modifying user/member roles. Prevents privilege escalationHighMicrosoftdeny
adm-settings-approveApprove settings changesRequire approval for any settings modificationHighGoogle, Microsoftrequire_approval
adm-member-add-approveApprove member additionsRequire approval before adding members to teams/groupsMediumMicrosoft, Telegramrequire_approval

Content Safety

Cross-cutting guards that apply to all outbound or inbound content regardless of action type.

Outbound (Request-Side)

Guard IDNameDescriptionRiskProvidersAction
cs-profanity-globalGlobal profanity filterBlock any outbound content containing profane language. Applies to all write operationsHighGoogle, Microsoft, Telegramdeny
cs-pii-globalGlobal PII guardFlag any outbound content containing PII (SSN, credit card, phone patterns). Applies to all write operationsHighGoogle, Microsoft, Telegramrequire_approval
cs-ip-address-guardIP address leak guardBlock outbound content containing internal IP addresses (10.x, 192.168.x, 172.16-31.x)MediumGoogle, Microsoft, Telegramdeny

Inbound (Response-Side)

Guard IDNameDescriptionRiskProvidersAction
cs-pii-redact-globalGlobal PII redactionRedact PII (email, phone, SSN, credit card, IP address) from all API responsesMediumGoogle, Microsoft, Telegramfilter

Summary by Action Type

Action TypeTotal Guardsdenyrequire_approvalfilter
Messaging13670
File Sharing7430
Calendar4040
Data Reading5004
Destructive7340
Administrative3120
Content Safety4211
Total4316215

Download

For the complete dataset with all fields, download the source CSV: Download full CSV