feat: Expose MCP gateway keepalive_interval in workflow config schema

[lpcox]

Summary

gh-aw-mcpg#3079 adds a keepalive_interval setting to the MCP gateway config that sends periodic pings to HTTP MCP backends, preventing session expiry during long-running agent tasks. This fixes the safeoutputs session timeout bug (gh-aw#23153).

However, there is currently no way for workflow authors to control this setting — it is a gateway-level config (GatewayConfig.keepalive_interval) that gh-aw passes via stdin JSON when launching the gateway. Workflow frontmatter has no corresponding field.

What needs to change

1. Add keepalive-interval to the gateway config schema that gh-aw generates when launching the MCP gateway, so it can be set per-repo or per-workflow

2. Expose it in the appropriate config layer — either:

   • A repo-level setting (e.g., in .github/aw/config.yml or similar) — most likely, since this is an infrastructure concern not a per-workflow concern
   • Or a workflow frontmatter field (e.g., under gateway: or mcp:) — if per-workflow control is desired

3. Pass it through to the stdin JSON config that gh-aw sends to the gateway on startup:

   {
     "gateway": {
       "keepaliveInterval": 1500
     }
   }

Accepted values (from gh-aw-mcpg#3079)

Value	Behavior
unset / 0	Default: 1500 seconds (25 minutes)
> 0	Custom keepalive interval in seconds
-1	Disable keepalive pings entirely

Note: 0 is treated as "unset" and silently defaulted to 1500 due to Go zero-value semantics in the gateway. A comment has been left on gh-aw-mcpg#3079 suggesting pointer semantics to make 0 mean "disabled" instead of requiring the -1 sentinel.

Use cases

• Long-running autoloop workflows (ML training, large builds): default 25-min keepalive is correct, no change needed
• Short-lived workflows with external session management: may want to disable keepalive (-1)
• Custom backends with different idle timeouts: may need shorter intervals (e.g., 300 seconds for a 10-min timeout)

References

• MCP gateway PR: fix: prevent safeoutputs session expiry during long-running agent tasks gh-aw-mcpg#3079
• Session expiry bug: MCP gateway drops out in long running jobs: Streamable HTTP error: Error POSTing to endpoint: session not found #23153
• Detailed reproduction: MCP gateway drops out in long running jobs: Streamable HTTP error: Error POSTing to endpoint: session not found #23153 (comment by @insop)