Configure Agent channels and credentials

These are the Channels and Credentials tabs of the Agent detail page. Changes on both tabs apply only after Deploy new version.

Channels tab

The Slack card shows Not configured, Disabled, or Configured (enabled) based on saved settings: Socket Mode, selected channels, and allowed users. This is not a live connection indicator.

Version cards are read-only. On the new version draft, Configure Slack or Edit Slack opens a drawer; Disable Slack saves a disabled channel setting. These changes affect future deployments, including other Agents sharing that Configuration. They do not stop a running channel or modify an existing version. Channels require Dedicated execution; unsupported native settings can make the simple editor unavailable.

Slack editor

Name search needs a reachable Slack directory; when it is unavailable, the drawer says so and exact IDs still work. Channel access and direct-message access are independent: channel people edit the selected channels' users. Switching to Open writes allowFrom: ["*"], Pairing or Allowlist write the listed people to allowFrom, and Disabled keeps the saved list. Save channel edits before changing credentials on the Credentials tab. A save keeps each Secret binding it does not change, so it needs Secret operate on that Secret, as described in Channel Secrets.

A Configuration save refuses an enabled channel whose token reference has no Secret binding. The refusal names the setting and the binding, for example channels.slack.botToken and SLACK_BOT_TOKEN, once, under the Configuration JSON when you edit it there or on Create Agent. Select the token Secrets in the Slack editor or on the Credentials tab, or remove the reference. A token must be exactly one ${NAME}: $NAME, a fallback, or other text, spaces or references with ${NAME} are refused, and the message names the form. The Slack editor opens tokens saved as ${SLACK_APP_TOKEN} and ${SLACK_BOT_TOKEN} and keeps that form. See Slack setup for credentials and policy details.

In any Secret picker, typing a Secret's exact name selects it when you press Enter or leave the field, as choosing its suggestion does; other text restores the bound Secret. Choose Refresh Secret list to find a Secret created in another tab or retry a failed list read. It preserves your search, focus, and staged choices without saving. The highlighted Secret stays highlighted by identity; if removed, choose again before pressing Enter. Refresh is unavailable while a read is pending; repeated activation is not queued. Existing access restrictions still apply.

Create Secret stores the value immediately. Cancelling the channel drawer discards token selections but does not delete that Namespace Secret. The modal never reads an existing value. If the name already exists in this Namespace, correct the Name and retry; both fields remain filled and the existing Secret is unchanged. See the Console reference for binding permissions and save behavior. Apply the saved draft with Deploy new version before expecting the running Agent to use it.

Microsoft Teams has no console editor. Existing Teams settings remain visible in native Configuration JSON, but a Teams-enabled draft cannot deploy through the console. Use the operator workflow for those Agents.

Credentials tab

Harness authentication

Authentication source determines how the harness gets model credentials:

Save authentication source saves the Agent binding for a future deployment. It needs Agent update, Configuration read, operate on each Secret and credential source that the Agent and its Configuration bind, and read on a ChatGPT service account it uses; without them, the source is read-only and a note names what is missing. For API keys and Service Accounts tokens, it also grants the Agent access to that exact Secret. Granting needs Installation administer. When a member saves the source, the binding saves, but the member's session sends no grant. The form stays editable and says that deploying checks the Agent's access: if the Agent cannot use the Secret, ask an Installation administrator to give the Agent operate on it, for example by opening the Agent's Credentials and selecting Save authentication source. Deploy new version stays available. When an administrator opens Credentials, the tab reads Namespace IAM, says when that grant is missing and holds Deploy new version until it is granted. When your access no longer includes updating this Agent, the warning asks an Installation administrator to grant it. If the grant was already in place, Reload authentication source clears the warning. If a grant request is refused or its outcome is lost, the console says so and offers Retry credential access. The retry checks the saved binding and does not resave it. If the binding changed, or the save outcome is unknown, use Reload authentication source first. Deploying needs Secret operate on each Secret the draft binds. When your access does not include it, Deploy new version is disabled and names those Secrets. A bound Secret you cannot read is not named: the button is disabled and asks for Secret read and operate on it. The Secret field says its name is unavailable. Other deployment authorization failures remain visible beside Deploy new version; check both your deployment permission and the Agent's credential access. Changing between API key and Service Accounts clears the selected Secret so a token is not silently reused for another authentication method. The account availability message describes discovery, not model readiness. See harness authentication.

Channel Secrets

This section appears when Slack is enabled in the saved Configuration.

The pickers bind only SLACK_APP_TOKEN for the app token and SLACK_BOT_TOKEN for the bot token, and only the keys the Slack tokens read. HTTP mode without an appToken reference reads no app token, so it shows only the Slack bot token picker; HTTP mode that keeps appToken: "${SLACK_APP_TOKEN}" shows both pickers. When the tokens read other keys, for example named accounts, other ${NAME} references or each other's key, the section lists each key as Bound or Missing without pickers. It also lists the other keys the Slack settings read, such as a signing secret's, below the pickers. Bind them through the Configuration API; Deploy new version waits until each is bound and the Agent can use its Secret. Deployment checks the Agent's access to every bound Secret, so a bound key no Slack setting reads, such as SLACK_APP_TOKEN left from Socket Mode, shows as Bound · not read by the Slack settings and needs that access too. Disable Slack keeps the bindings, so with Slack off the section still lists each bound Secret whose access is not confirmed. While an Installation administrator's console reads that access, Deploy new version waits, and the save message names deployment as the next step only after the Agent's access is confirmed. Until then, Deploy new version names the token or key whose Secret waits and says to select Retry Secret access in Credentials. Retry Secret access stays available for all of them. When your access no longer includes editing this Configuration, the panel offers no Retry and the warning asks an Installation administrator to grant or confirm the access. If the API refuses a save for a channel setting, the panel shows the API's message, which names the setting. Fix that setting in the Configuration JSON on the Configuration tab. If the message asks for a Secret binding of another name, bind that name through the Configuration API.

First-deployment credential generation requires Agent read and operate when generated credentials are missing; deployment also requires deploy. Saving channel Secrets additionally requires Secret, Configuration, and Namespace IAM permissions. Every Configuration save needs Secret operate on each Secret it keeps bound, and Save channel Secrets keeps every binding it does not change. When the Secret you cannot operate is a Slack token, the panel names it: choose another Secret for it, or the console refuses the save before sending it. Disable Slack stays disabled until you do. Once that save succeeds, the plugin, repository and authentication source controls become editable without a reload. When it is bound under another key, the pickers are read-only and name that key. These are multiple writes, so a failure can leave partial progress. Granting the Agent a Secret needs Installation administer: when a member saves a token, the Configuration saves, but the member's session sends no grant, so it does not verify the Agent's access. The console asks an Installation administrator to open the Agent's Credentials and select Retry Secret access, and offers no retry itself. Deploy new version stays available, and Reload draft clears the warning: the member's console cannot read the grant, so the API refuses a deployment it does not cover. Outcome unknown means refresh and inspect saved state before retrying. Save requires at least one changed selection and a bound Secret for each token. Changing the picker switches the referenced Secret; it does not overwrite an existing shared Secret value. Existing values are never fetched or displayed.