fix(platform-agent): ship management-MCP diagnostic in heartbeat (cp#3164) #171
Reference in New Issue
Block a user
Delete Branch "fix/3164-platform-mcp-diag-observability"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Closes the #3164 observability blind spot: the management-MCP failure (why the molecule-platform MCP server fails to start on fresh platform agents → status=failed → concierge can't create_workspace → staging E2E red) only logged to container stdout, invisible on locked-down prod boxes. identity_gate_payload now ships platform_mcp_diag {on_platform_agent_image, mcp_binary_present, mcp_settings_entry, mcp_command_resolved} so the CP (which knows kind=platform) can record the cause without box SSH. Tests: 40 identity + 89 heartbeat/register green. Companion follow-up: CP records platform_mcp_diag when degrading a kind=platform agent. — devops-engineer / CEO-Asst
APPROVED on head
c346073b.5-axis review:
identity_gate_payload()now always includesplatform_mcp_diagwith the four requested signals. The fields map to the existing platform-agent image marker, legacy baked MCP path, Claude settings entry, andmolecule-platform-mcpPATH resolution. This is additive observability and does not change the existingmcp_server_presentgate decision or loaded-tools behavior._settings_has_management_mcp(). Missing binary/settings/PATH resolve to booleans/None rather than exceptions.shutil.which()on heartbeat/register cadence is acceptable for diagnostic telemetry.APPROVE runtime#171 @c346073b.
Five-axis review:
No blocking findings.