Bundled Slack Channel Driver

The bundled SlackChannelDriver reads people and channels for the Console's name picker. It uses the Agent's selected same-Namespace SLACK_BOT_TOKEN Secret; the controller does not configure or store another token.

The Driver calls Slack auth.test to require a bot identity and identify its workspace, then users.list or conversations.list. The bot needs users:read for people, channels:read for public channels, and groups:read for private channels. Slack returns only resources visible to the bot. The Driver scans up to three provider pages for a nonempty search query and one page without a query. It returns a cursor when more pages remain. The Console displays the workspace and saves selected IDs, never names. An empty, null, or absent provider cursor marks the final page, as defined by Slack pagination.

For exact-ID searches and saved IDs, the Driver calls users.info or conversations.info directly. The Console shows selected names as removable chips, with exact IDs available on hover. IDs the bot cannot see remain visible without a name. Search in the field and choose a result, or paste exact IDs and press Enter. The Console waits 300 ms after typing and shows up to five results per page. Enter searches immediately. Next and Previous reuse fetched matches before requesting another provider page. Unselected search text is never saved. Channel access explicitly selects Specific people or Everyone in these channels; removing the last person does not switch to everyone.

An invalid token, missing scope, rate limit, invalid response, or unavailable Slack service produces a safe error without returning the token or upstream payload. Non-success HTTP responses release their unused response body before returning the error. Retry after fixing the token or scopes. Exact IDs can be entered when directory browsing is unavailable.

Credential validation

Before API provisioning or deployment, enabled Slack accounts require Secret-backed native environment references: an env SecretRef, or a token that is exactly one ${NAME}, with NAME bound to a Secret. Socket Mode app tokens must start with xapp-; bot tokens must start with xoxb- and pass auth.test with a bot identity. An explicit Enterprise Grid installation requires an enterprise ID; other installations require a workspace ID. User and channel lookup still require a workspace ID. HTTP mode skips the app-token check. Disabled Slack accounts make no provider call. Validation has an eight-second request budget and never opens a Socket Mode connection or sends a message.

Failures return sanitized CHANNEL_CREDENTIAL_* errors with the native field path. A retryable 503 means Slack or its route failed. When the production API has no proxy URL, its egress is closed (OCC_API_HTTPS_EGRESS_OPEN is not true), and auth.test gets no answer from Slack (connection, DNS, or time budget failure), validation instead returns 409 RESOURCE_CONFLICT. Helm sets OCC_API_HTTPS_EGRESS_OPEN to true when Google or OIDC sign-in is enabled with empty egressCidrs, because that rule allows TCP 443 to any address. The message names the proxy settings and says to try again if the API has another approved route to Slack, because an open route that timed out gets the same answer. App-token prefixes do not prove validity, scopes, or app/bot pairing. Validation checks the current values; credentials can change before runtime startup, which remains a separate connectivity check.

Enable lookup in production

The production API selects the Slack Driver. With default-deny egress, Slack lookup and credential validation need a proxy. With the recommended Helm setting slackProxy.enabled: true, the chart:

• creates a private openclaw-enterprise-slack-proxy Service;
• points OCC_CHANNEL_DIRECTORY_PROXY_URL at that Service DNS name;
• allows API Pod egress only to the proxy Pod selector; and
• allows the proxy public IPv4 egress on TCP 443, excluding private and reserved ranges.

The bundled proxy accepts HTTP CONNECT only for Slack hostnames on port 443. The API accepts a DNS proxy URL only when Helm also sets the matching OCC_CHANNEL_DIRECTORY_MANAGED_PROXY_HOST; it rejects any other DNS proxy URL.

The bundled proxy closes a tunnel if it has pending writes to the client for 10 continuous seconds while the upstream connection is open. A quiet Socket Mode tunnel can stay open when no writes are pending.

Operators can instead set Helm api.channelDirectoryProxyUrl to an approved external HTTP or HTTPS proxy, such as http://198.51.100.25:3128 with the example IP and port replaced. The value must contain one literal IPv4 address other than 0.0.0.0 and an explicit port, without credentials or a path. The chart passes it to the API as OCC_CHANNEL_DIRECTORY_PROXY_URL and allows API Pod egress only to that IP and port. The proxy must permit CONNECT slack.com:443; restrict its other destinations at the proxy. The chart does not grant the worker or Agent Pods this egress.

The Driver sends Slack API requests through the selected tunnel and verifies Slack's TLS certificate. Keep the selected bot token in the same-Namespace Secret; Helm values need no token or other credential for the proxy. Without a reachable Slack route, lookup returns unavailable, and credential validation answers as described in Credential validation. Exact-ID entry does not bypass credential validation. See the production controller settings for the environment contract.

See the ChannelDriver contract for OCC authorization and Slack Configuration for gateway token bindings.